Analyzing Social Media Networks with NodeXL: Insights from a Connected World [2 ed.] 012817756X, 9780128177563

Table of contents :
Cover
ANALYZING SOCIAL MEDIA NETWORKS WITH NODEXL: Insights from a Connected World
Copyright
Dedication
About the authors
Contributors
Preface
Acknowledgments
Part I Getting started with analyzing social media networks
1
Introduction to social media and social networks
Introduction
A historical perspective
The rise of enterprise social media applications
Individual contributions generate public wealth and risks
Who should read this book
Applying social media to national priorities
Worldwide efforts
Practitioner’s summary
Researcher’s agenda
References
Additional resources
2
Social media: New technologies of collaboration
Introduction
Social media defined
Social media design framework
Size of producer and consumer population
Pace of interaction
Genre of basic elements
Control of basic elements
Types of connections
Retention of content
Social media examples
Asynchronous threaded conversation
Email
Email lists, discussion forums, Reddit, Quora, and Q&A sites
Synchronous conversation
Chat, instant messaging, and texting
Audio and video conferencing
The World Wide Web
Collaborative authoring
Wikis
Shared documents
Blogs and podcasts
Microblogs and activity streams
Multimedia blogs, podcasts, and livestreams
Social sharing
Video and TV
Photo, images, and art
Music
Bookmarks, news, and books
Social networking services
Social and dating
Professional
Niche networks
Online markets and production
Financial transactions
User-generated products
Review sites
Idea generation
Games and virtual worlds
Virtual reality worlds
Massively multiplayer games
Mobile services
Location and augmented reality apps and games
Practitioner’s summary
Researcher’s agenda
References
Additional resources
3
Social network analysis: Measuring, mapping, and modeling collections of connections
Introduction
The network perspective
A simple Twitter network example
Vertices
Edges
Network data representations
Types of networks
Egocentric, partial, and full networks
Unimodal, multimodal, and affiliation networks
Multiplex networks
The network analysis research and practitioner landscape
Network analysis metrics
Aggregate network metrics
Vertex-specific network metrics
Degree centrality
Betweenness centrality: Bridge scores for boundary spanners
Closeness centrality: Distance scores for strategically located people
Eigenvector and PageRank centrality: Influence scores for strategically connected people
Clustering coefficient: How connected are my friends?
Grouping, clustering, and community detection algorithms
Structures, network motifs, and social roles
Social networks in the era of abundant computation
The era of abundant social networks: From the desktop to your hand
Tools for network analysis
Node-link diagrams: Visually mapping social networks
Common network analysis questions applied to social media
Practitioner’s summary
Researcher’s agenda
References
Additional resources
Part II NodeXL tutorial: Learning by doing
4 Installation, orientation, and layout
Introduction
Downloading and installing NodeXL
Getting started with NodeXL
Opening a new NodeXL file
NodeXL menu ribbon
Spreadsheet and graph pane
Manually entering data
Importing data
Showing the graph
Highlighting an edge or vertex
Resizing and moving the graph pane
Layout: Arranging vertices in the graph pane
Manual layout
Automatic layout
Adjusting Fruchterman-Reingold settings
Updating the graph pane
Preserving a layout
Graph pane tools
Graph pane options
Undirected and directed graph type
Changing the type of network
Reciprocated edges
Working with NodeXL files
Saving NodeXL files
Exporting an existing NodeXL file
Opening an existing NodeXL file
Opening a NodeXL file created on another computer
Creating a trusted location for NodeXL files from the Internet
Practitioner’s summary
Researcher’s agenda
References
NodeXL papers
5
Labeling and visual attributes
Introduction
Labeling
Viewing attribute data in the ABCD network file
Labeling vertices
Adding tooltips
Formatting, positioning, and truncating labels using label options
Label vertex shape
Labeling edges
Visual properties
Vertex color
Vertex shape
Vertex size
Vertex opacity
Vertex visibility
Edge visual properties
Showing the graph legend
Saving graph images and right-click graph menu
Graph options
Practitioner's summary
Researcher’s agenda
References
6
Calculating and visualizing network metrics
Introduction
ABCD network example
Computing graph metrics
Overall graph metrics
Vertex-specific metrics
Degree
Betweenness centrality
Closeness centrality
Eigenvector centrality
PageRank
Clustering coefficient
Marvel cinematic universe network example
Visualizing and interpreting metrics in a bimodal network
Mapping graph metrics to X and Y coordinates
CSCW 2018 conference Twitter network example
Calculating and interpreting directed network metrics
Examining top items output
Examining time series output
Practitioner’s summary
Researcher’s agenda
References
7
Grouping and filtering
Introduction
U.S. Senate voting analysis
Filtering edges to identify groups within a network
Using dynamic filters
Creating groups based on vertex attribute
CSCW 2018 Twitter network analysis
Filtering out self-loops using the edge visibility column
Grouping and visualizing connected components
Using dynamic filters to filter based on time
Filtering based on vertex metrics and the visibility column
Automatically identifying groups based on network clustering algorithms
Group properties and metrics
Group layout and labels
Creating subgraph images
Federal Communications Commission (FCC) lobbying coalition network
Practitioner’s summary
Researcher’s agenda
References
Additional resources
8
Semantic networks
Introduction
Creating the Twitter Gardasil HPV word pair network
Calculate word and word pair metrics
Iteratively refine the list of skipped words
Creating a new word to word network file
Analyzing word networks
Examining vertex and edge metrics
Examining data by groups
Visualizing work networks
Visualizing computing dissertation and thesis connections
Practitioner’s summary
Researcher’s agenda
References
Suggested reading
Part III Social media network analysis case studies
9
Email: The lifeblood of modern communication
Introduction
History and definition of email
Email networks
What questions can be answered by analyzing email networks?
Personal email network questions
Organizational email network questions
Working with email data
Preparing email
Importing email networks into NodeXL
Cleaning email data in NodeXL
Remove duplicate email addresses for the same individual
Count and merge duplicate edges
Analyzing personal email networks
Creating an email overview visualization
Step 1: Import data into NodeXL
Step 2: Clean data
Step 3: Filter data
Step 4: Compute graph metrics and add new columns
Step 5: Visualize the email social network
Step 6: Understand social network visualizations and metrics data
Creating an expertise network email graph
Step 1: Import email social network data into NodeXL
Step 2: Clean data
Step 3: Compute graph metrics and add new columns
Step 4: Filter data
Step 5: Visualize network
Step 6: Understanding the network visualization and data
Creating a living org-chart with an organizational email network
TechABC's organizational unit email network
Normalizing and filtering TechABC's data
Creating an overview of TechABC's communication patterns
Examining TechABC's research division
Historical and legal analysis of Enron email
Identifying key individuals using content networks
Practitioner's summary
Researcher's agenda
References
10
Thread networks: Mapping message boards and email lists
Introduction
Definition and history of threaded conversation
What questions can be asked
Threaded conversation networks
Identifying important people and social roles in the CSS-D Q&A reply network
Understanding groups at Ravelry
Practitioner's summary
Researcher's agenda
References
Further reading
11
Twitter: Information flows, influencers, and organic communities
Introduction
Defining your topic-networks: Formulating a social media monitoring query
Twitter data collection
The raw data layout
Network analysis
Vertex-level metrics
In and out degree centrality
Betweenness centrality
User reciprocity
Network-level metrics
Overall metrics
Graph density
Graph reciprocity
Groups
Visualization
User-level visual properties
Cluster-level layout and visual properties
Analysis of content
Share your work on the NodeXL graph gallery
Practitioner's summary
Researcher's agenda
References
Suggested reading
12
Facebook: Public pages and inter-organizational networks
Introduction to Facebook: The social graph of 2 billion people
Facebook networks
Facebook API limits
Organizational networks: Fan pages
Preparing the data collection
Data collection
Getting to know the data
The Edges Worksheet
The Vertex Worksheet
Network analysis
Vertex-level analysis
Cluster-level analysis
Visualization
Practitioner’s summary
Researcher’s agenda
References
Suggested reading
13
YouTube: Exploring video networks
Introduction
What is YouTube?
YouTube’s structure
Videos
The user channel
Networks in YouTube
Video networks
Users’ networks
Hubs, groups, and layers: What questions can social network analysis of YouTube answer?
Video networks
User networks
Importing YouTube data into NodeXL
Importing video data
Importing user data
Ethical considerations
Problems with YouTube network data
Preparing YouTube network data
Analyzing YouTube networks
User networks
Video networks
The YouTube “makeup” video network
Practitioner’s summary
Researcher’s agenda
References
Suggested reading
14
Wiki networks: Connections of culture and collaboration
Introduction
Key features of wiki systems
Wiki networks from edit activity
Wiki networks of general interest
Using the NodeXL MediaWiki page network importer to access Wikipedia networks
Understanding topics through page-to-page connections
Data collection and processing
Identifying key topics across Wikipedia language communities
Analyzing the structure of discussion page interaction
Mapping networks and identifying disputes within the English International Whaling Commission talk page
Identifying productive members of a talk page community
Choosing the right sample frame for your wiki research
Practitioner’s summary
Researcher’s agenda
References
Index
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
Y
Back Cover

Citation preview

ANALYZING SOCIAL MEDIA NETWORKS WITH NODEXL Insights from a Connected World

ANALYZING SOCIAL MEDIA NETWORKS WITH NODEXL Insights from a Connected World SECOND EDITION Derek L. Hansen Ben Shneiderman Marc A. Smith Itai Himelboim

Morgan Kaufmann is an imprint of Elsevier 50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States © 2020 Elsevier Inc. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions. This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein). Notices Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary. Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility. To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein. Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-12-817756-3 For information on all Morgan Kaufmann publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Katey Birtcher Acquisition Editor: Steve Merken Editorial Project Manager: Ali Afzal-Khan Production Project Manager: Punithavathy Govindaradjane Cover Designer: Miles Hitchen Typeset by SPi Global, India

Dedication Derek: To Maren Hansen Itai: To Jonathan and Nadav Ben: To Jennifer Preece Marc: To my strong ties to Madeline, Eli, and Christine

v

About the Authors Derek L. Hansen (http://www.et.byu.edu/~dlh24/) is an associate professor of Information Technology & Cybersecurity at Brigham Young University. His research and teaching focus on understanding and designing social technologies, tools, and games for the public good in areas ranging from education to citizen science to fitness to transcription of historical documents. He has received over $2 million in grants to help develop and evaluate novel technical interventions, alternate reality games, and educational simulations (called Playable Case Studies) with highly talented faculty, students, and professionals from a variety of disciplines. He completed his Ph.D. from the University of Michigan’s School of Information where he was a National Science Foundation-funded interdisciplinary STIET Fellow focused on understanding and designing effective online sociotechnical systems. Ben Shneiderman (www.cs.umd.edu/~ben) is an emeritus distinguished university professor in the Department of Computer Science and founding director (1983–2000) of the Human-Computer Interaction Laboratory (www.cs.umd.edu/hcil) at the University of Maryland. He was elected as a fellow of the Association for Computing (ACM), the American Association for the Advancement of Science (AAAS), the IEEE, and the National Academy of Inventors. He received the ACM SIGCHI lifetime achievement award and was elected as a member of National Academy of Engineering. He is the lead author of Designing the user interface: Strategies for effective human-computer interaction (6th ed., 2016). He wrote Leonardo’s laptop: Human values and the new computing technologies and the new ABCs of research: Achieving breakthrough collaborations. Marc A. Smith is a sociologist specializing in the social organization of online communities and computermediated interaction. He leads the Connected Action consulting group and lives and works in Silicon Valley, California. He is the coeditor, with Peter Kollock, of Communities in cyberspace (Routledge), a collection of essays exploring the ways identity, interaction, and social order develop in online groups. His research ­focuses on

computer-mediated collective action: the ways group dynamics change when they take place in and through social cyberspaces. Many “groups” in cyberspace produce public goods and organize t­ hemselves in the form of a commons (for related papers see http://connectedaction.net/marc-smith). His goal is to visualize social cyberspaces, mapping and measuring their structure, dynamics, and life cycles. He also oversees the Social Media Research Foundation (https://www.smrfoundation.org/), which is a nonprofit organization that oversees the ongoing development of NodeXL Pro (https://nodexlgraphgallery.org/) and its use in research and teaching. The Connected Action consulting group (http://www.connectedaction.net) applies social science methods in general and social network analysis techniques in particular to enterprise and Internet social media usage. He received a B.S. in International Area Studies from Drexel University in Philadelphia in 1988, an M. Phil. in social theory from Cambridge University in 1990, and a Ph.D. in Sociology from UCLA in 2001. He is a member of the Media-X research consortium at Stanford University. Itai Himelboim is an associate professor at the University of Georgia’s Grady College of Journalism and Mass Communication, Department of Advertising and Public Relations and a director of the SEE Suite—the Social Media Engagement and Evaluation lab (http:// seesuite.uga.edu). He completed his Ph.D. from the University of Minnesota’s School of Journalism and Mass Communication, focusing on the intersection between civil society and computer-mediated social networks. His research and teaching interests include social network analysis of large social media data related to news, brands, politics, health, and international affairs. In his research, he examines the network structures that emerge when users interact on Twitter and other social media spaces, patterns of information diffusion, the emergence of network clusters as information echo chambers, key information sources in these networks, as well as identifying key users and content that bridges these information silos.

xi

Contributors

Md. Mahbub Or Rahman Bhuyan  Department of Sociology and Anthropology, Ohio University, Athens, OH, United States

Bryan M. Trude  Department of Advertising & Public Relations, Grady College of Journalism and Mass Communication, University of Georgia, Athens, Georgia

Nina Cesare  School of Public Health, Boston University, Boston, MA, United States

Howard T. Welser  Department of Sociology and Anthropology, Ohio University, Athens, OH, United States

Jen Golbeck  College of Information Studies, University of Maryland, College Park, MD, United States

xiii

Preface

We live in the era of networks. Occasionally the aspirations of academic researchers are in harmony with the needs of software developers, entrepreneurs, and government agency staffers. In our case, the authors brought together complementary backgrounds in information studies, communications, computer science, and sociology, as well as a shared interest in the interdisciplinary topics of human-computer interaction, network analysis, social media, and information visualization. Together, we have worked to build tools that broaden access to insights into complex connected network structures. Networks are a topic that is gaining interest from the growing community of entrepreneurs who are coping with the success of social media commercial platforms such as Twitter, Facebook, YouTube, and the equally ­remarkable open source communities that produce valuable resources such as Wikipedia. Social media is now perceived of as a collection of complex networks which can be best understood by drawing on social science methods designed to help identify connections, influencers, and social roles. Tools for social media network analysis and visualization have been emerging from many research groups and startup companies. These pioneering network analysis tools often require programming skills and knowledge of technical network terminology, making it a challenge for those without programming skills to import and make sense of network data. Today measurements of networks are a mature topic, but research questions remain on the best way to layout and cluster networks with algorithms. Network innovation has expanded dramatically in the past decades, producing breakthroughs that expanded the scale and level of insight into networks that are possible to capture. Similarly, strategies for filtering, visualizing, and decluttering networks have matured as more users tackle a broader variety of problems with increasingly large networks. The authors have been fortunate to be part of a team with unique skills that continue to develop the NodeXL tool. The project was originally funded by Microsoft Research and for over 10  years it has been sponsored by the Social Media Research Foundation (http://smrfoundation.org), a nonprofit organization dedicated to the development of NodeXL and related tools. The SMR Foundation is focused on supporting scholarship related to social media via tool building and research assistance.

Contributors to the foundation are distributed across the United States, Europe, and Asia with links to others around the world. Our members have created plugins and features for extending NodeXL, adding support for language and content analysis, better visualizations, and better reports and presentations. NodeXL has continued to support the academic community, who have used the tool to publish over 7000 articles citing NodeXL or the first version of this book. Scholars from dozens of disciplines ranging from computer science to communications to digital humanities to history have used NodeXL to tell stories about the networks that underlie our society. They have also found NodeXL to be an invaluable tool in teaching social network analysis to undergraduate and graduate students across the world. Its integration with the familiar Excel spreadsheet program, built-in importers from social media sites, and rich analysis and visualization features make it ideal for those starting to learn social network analysis. Meanwhile, NodeXL has transitioned from a primarily academic project into a viable commercial tool, as companies have turned to NodeXL to understand how social media users are engaging with their brand, products, and competitors. There have been hundreds of thousands of downloads of NodeXL since its inception. Companies use NodeXL to identify thought leaders in new markets and recognize how they fit into the larger social media landscape. The ongoing success of NodeXL has allowed it to continue developing into an increasingly sophisticated network and content analysis and visualization tool that fills an important niche in the social network analysis landscape. The future of NodeXL looks bright, with a highly active research and development community that spans the globe, a growing number of users, and an excellent core team of developers. An increasing number of YouTube videos, online tutorials, and supplemental materials are becoming available; though this book is by far the most comprehensive introduction to NodeXL available. Updates continue to be pushed at least monthly, and it is spotlighted at numerous workshops and conference presentations each year. New features and developments are designed to meet the needs of the NodeXL community members and questions or bugs are quickly resolved when uncovered.

xv

xvi

PREFACE

The success of NodeXL has validated our initial goal of making social network analysis, especially of social media data, more accessible to the masses. Constant feedback from researchers and practitioners using NodeXL has helped us improve usability, envision new features, and raised our aspirations about what we and our users could accomplish in terms of gaining

actionable insights from social media analytics. Although we are pleased with what NodeXL users are able to accomplish, we are humbled by the richness and diversity of social media analytical possibilities. The opportunities and challenges are substantial, which will keep researchers and developers productively engaged for many years to come.

Acknowledgments

The authors would like to thank the many people who have made this document and the NodeXL project possible. First, the core members of the NodeXL design and development team include Harald Meier and Arber Ceni on whom we depend for their remarkable programming prowess, thoughtful discussions about features, and always courteous help to us and the NodeXL user community. In addition, Natasa MilicFrayling from the University of Nottingham, Eduarda Mendes Rodrigues from the UK Mobile Phone company Giff Gaff, Cody Dunne at Northeastern University, Jana Diesner at University of Illinois, Urbana Champaign, and Jure Leskovec at Stanford University all have made significant contributions to the NodeXL project. We are grateful also to Adam Perer from IBM Research for his intellectual contributions to our grand adventure. We remain in debt to Tony Capone for his software development work in the initial years of NodeXL. We also thank Emily Mason, Chad Doran, and Rachel Collins, who collected datasets used in the book as part of their coursework and came up with compelling analyses of them. Special thanks to Chris Wilson of Slate Magazine for sharing the Senate 2007 voting data and Jared Stewart for creating a comparable datasets for more recent years. The students of several classes who were assigned projects with NodeXL have been patient and forgiving as we refined the rough edges. Many hundreds of thousands of people have downloaded NodeXL, and many have created research and business results using the tool. We appreciate the time and attention our users give the tool and the project and hope they will continue to upgrade with us as the project grows. We are grateful to these and many other people for their efforts to make NodeXL an easy and useful tool for understanding complex networks. Ai Addyson-Zhang, Stockton University Hayan Ajjan, Elon University Julie Albright, University of Southern California Wasim Ahmed, University of Northumbria Harith Alani, Open University Nasir Asar, Highpoint University Diana Asher, University of California, Los Angeles Brandy Aven, Carnegie Mellon University Vladimir Barash, Grafika George Barnett, University of California at Davis Daniel Bassill, TutorMentor

Sue Beckingham, Sheffield Hallym University Steve Boland, Next in Non-Profits Kirk D Borne, Booz Allen Hamilton Katy Borner, Indiana University Aras Bozkurt, Anadolu University Marie Brugere, Social media expert Julian Chin, University of Illinois, Urbana Champaign Diane Cline, George Washington University Noshir Contractor, Northwestern University James Cook, University of Maine, Augusta Rob Cross, University of Virginia Helen Darling, Sumfood Wayne deFremery, Sogang University Scott Dempwolf, University of Maryland Daniel Erasmus, NewsConsole Randy Farmer, online community expert Liam Farrell, Author Kyle Findlay, Kantar Analytics Lise Getoor, University of California, Santa Cruz Scott Golder, Capital One Ian Griffin, Executive Communications Jacob Groshek, Boston University Libby Hemphill, University of Michigan Bernie Hogan, Oxford University, Oxford Internet Institute Bill Johnston, Structure3C David Kaplan, Global Investigative Journalism Network Brian Keegan, University of Colorado John Kelly, Grafika Gohar Khan, Waikato University Leo Kim, Ars Praxia, Korea Valdis Krebs, Orgnet Cliff Lampe, University of Michigan Jeremy Harris Lipschultz, University of Nebraska, Omaha Graham Mackenzie, Scottish National Health Service Surgeon Sorin Matei, Purdue University Fil Menzcer, Indiana University Anne Merick, The Bravo Group Nasri Messarra, Saint Joseph University Luisa Milic, Ideya.eu Aldo de Moor, Online Community Expert Scott Moore, Online Community Expert Miriam Notten, LaRed Consulting Katherine Ognyanova, Rutgers University

xvii

xviii

ACKNOWLEDGMENTS

Niall O Gribin, Digital Destiny Marketing Mark Outhwaite, Outhentics: Healthcare research Han Woo Park, Yeungnam University Leysia Palen, University of Colorado Katy Pearce, University of Washington Ann Pegoraro, Laurentian University Andrew Pitts, Polinode Mike Quindazi, Price Waterhouse Coopers Lee Rainie, Pew Research Raquel Recuero, Universidade Federal de Pelotas Howard Rheingold, online community author and teacher Jason Schulman, Stockton University Kate Starbird, University of Washington Elisabeth Stedman, Smart Kiwis Jeremy Teoh, Chinese University of Hong Kong David Tindall, University of British Columbia Leslie Tkach-Kawasaki, University of Tsukuba

Zeynep Tufekci, University of North Carolina Rusty Tunard, Tufts University Diana Turecek, Middle East Broadcast Network Jeff Ubois, Macarthur Foundation Glen Waddell, University of Nevada, Reno Barry Wellman, University of Toronto Brooke Foucault Wells, Northeastern University James Witte, George Mason University Kara Wood, Succinct Social Media Scott Wright, University of Melbourne Gi Woong Yun, University of Reno Lyra Zeigler, Graphic Designer Ton Zijlstra, Online Community Expert In Memory Walter Pike Dana Rotman

P A R T

I

Getting started with analyzing social media networks This volume is organized in the form of a tree with roots, a trunk, and branches. The roots (Part I: Chapters 1 through 3) provide grounding in the history and core concepts of social media and social network analysis. The trunk (Part II: Chapters 4 through 8) focuses on the practical details of operating the free and open source NodeXL extension of the familiar Microsoft Excel spreadsheet application used for all exercises in this volume. And the branches (Part III: Chapters 9 through 14) each focus on one form of social media by describing each system, the nature of the networks that are created when people interact through it, and the kinds of analysis that can be performed to identify key people, documents, groups, and events. The results are actionable insights that can guide community managers, marketers, organi-

zations, and members as they try to improve the quality and value of their social media initiatives. Part I discusses the novel ways people connect through social media tools and the formal network analysis techniques that can elucidate those connections. Chapter 1 provides a high-level overview of social media initiatives and network science, explaining why the two can be so effectively combined. Chapter 2 describes the design space of social media tools, providing examples of different types of connections created by popular tools. Finally, Chapter 3 introduces the core concepts of social network analysis and visualization, which will be put to work in the remainder of the book to understand social media networks. The chapters assume no prior knowledge of these topics.

C H A P T E R

1 Introduction to social media and social networks O U T L I N E 1.1 Introduction

3

1.6 Applying social media to national priorities

7

1.2 A historical perspective

4

1.7 Worldwide efforts

8

1.3 The rise of enterprise social media applications

1.8 Practitioner’s summary

9

5

1.9 Researcher’s agenda

9

1.4 Individual contributions generate public wealth and risks

5

References

9

1.5 Who should read this book

6

Additional resources

1.1 Introduction

As these webs of content and connection grow, so does their individual, social, political and economic impact. Social media networks are increasingly the way we see and know about the world around us. The profound ways social media are changing society call for better tools and research to understand and document the different kinds of social media formations and the critical events that can hit these collective systems. To create better social media environments we will need to better understand the dynamics and patterns of existing platforms. Sailing the seas of social media has allowed for new connections and movements, but not all of them have been positive. Since a rising tide lifts all boats, the power of social media to amplify collective action has led to the revitalization of violent and divisive organizations, as well as those seeking knowledge, understanding and peaceful cooperation. The need for new tools to collect, analyze, visualize, and generate insights from these collections of connections is growing. In the same way that financial markets require accounting and auditing for proper operation, social media platforms are markets for ideas and attention that currently lack widespread tools for collecting and evaluating content. Left to their own devices, these markets of billions of messages, links, posts, edits, uploaded photos and videos, reviews, and recommendations seem

Billions of people create trillions of connections through social media each day, but few of us consider how each tap, swipe, click and keypress builds relationships that, in aggregate, form vast social networks. Using these social networks people often collectively create assets of significant value. Passionate users of social media tools such as email, blogs, microblogs, messenger systems and wikis eagerly send personal or public messages, post strongly felt opinions, or contribute to community knowledge to develop partnerships, promote cultural heritage, and advance development. Encyclopedias, operating systems, books, currencies, sports leagues and social movements have all been collectively created via social media tools. Devoted social networkers create and share digital media and rate or recommend resources to pool their experiences, provide help for neighbors and colleagues, and express their creativity. In other cases individuals, groups, companies, parties and nations use these tools to attack, mob, or confuse these collective assets and prominent individuals. The results are vast, complex networks of connections that link people to other people, documents, locations, concepts, and other objects, not all of which are valid or humane.

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00001-7

10

3

© 2020 Elsevier Inc. All rights reserved.

4

1.  Introduction to social media and social networks

to generate benign or even surprisingly positive results, though digging deeper can uncover concerted collusion and well funded manipulation. Like markets for stocks and other high risk assets, a variety of accounting and auditing tools and practices are needed to ensure a well run and sufficiently honest marketplace. As social media have emerged as a widespread platform for human interaction, the invisible ties that have always linked each of us to others have become more visible and machine readable. The result is a new opportunity to map social networks in detail and scale never before seen. The complex structures that emerge from webs of social relationships can now be studied with computer programs that create graphical maps of these connections. These tools integrate methods from the study of social networks and content analysis to capture the shape of the virtual crowd, highlighting the leading topics within them, and identify the people who occupy the key locations within these landscapes of ties and links. Creating maps of a wide range of social media activity could be helpful. Mapping the positive examples of social media enabled collective action may guide us to best practices for cultivating the positive outcomes, while mapping examples of conflicts and manipulation can help develop the tools to detect, deflect and deter malfeasance and manipulation of marketplaces of ideas. Computer information networks give transport to valid and invalid information equally, and the current design of social media platforms may even amplify the sensational and emotional over factual and reasonable information. News mixes with fiction via computer information networks in ways that allow many people to confuse the one for the other. Like meteorologists monitoring the daily variations of weather while keeping a watchful eye out for storms, social media analysts may be increasingly called to map and monitor the landscape of social media to ensure the cultivation of the best authentic discussions. Social network platforms have removed many traditional gatekeepers from prior eras of mass media, but the resulting marketplace of ideas is both more diverse and less able to distinguish facts from the beliefs of groups. Maps of social media networks can guide new journeys through the landscapes of previously uncharted connected conversations, finding the valuable assets while avoiding the pitfalls.

1.2  A historical perspective Network science focuses on the study of patterns of connection in a wide range of physical and social phenomena. Network researchers have explored foundational physical systems created by chemical and genetic connections, webs of consumption in which some animals eat others, and profound distributed human social

phenomena such as collective action, empathy, social cohesion, privacy, responsibility, markets, motivation, and trust. In the past few decades, network researchers have developed new data collection methods, innovative mathematical techniques, and surprising predictive theories. Just as Lord Kelvin (1824–1907) encouraged careful measurement as the method of advancing science, the new sciences of collective action, collaboration, and productive communities require new forms of measurement. Similarly, where Newton (1643–1727) and Leibniz (1646–1716) created the mathematical methods of calculus to grasp the physical world of objects in motion, social scientists are developing advanced mathematical methods for capturing social network evolution, diffusion, and decay. Like Galileo’s telescope (1564–1642), Hooke’s microscope (1635–1703), or Roentgen’s (1845–1923) X-rays, new information analysis tools are creating visualizations of never before seen social structures. Jupiter’s moon, plant cells, and the skeletons of living creatures were all revealed by previous technologies. Today, new network science concepts and analysis tools are making isolated groups, influential participants, and community structures visible in ways never before possible. Social network analysis is the application of the broader field of network science to the study of human relationships and connections. Social networks are primordial; they have a history that long predates systems like Facebook and WeChat, or even the first email message. Ever since anyone exchanged help with anyone else, social networks have existed, even if they were mostly invisible. Social networks are created from any collection of connections among a group of people and things. Social network science is itself relatively new, with roots in the early 20th century pushed forward by authors such as Jacob Moreno, Georg Simmel, and Linton Freeman. This in turn, built on two centuries of work in the mathematics of graphs and topology, also known as graph theory, developed by mathematicians such as Leonhard Euler. In the 21st century, network science has blossomed alongside a new global culture of commonplace networked communications. Billions of people use desktops, laptops, tablets, phones, and other devices to routinely create rich digital artifacts that they share with individuals and groups sometimes as large as many millions. Widespread network connectivity has arrived in just the past few decades, and now billions of people have changed their lives by creatively using social media. We use social media to bring our families and friends closer together, reach out to neighbors and colleagues, and invigorate markets for products and services. Social media are used to create connections that can bind local regions and span continents. These connections range from the trivial to the most valued, potent collaborations, relationships, and communities. Social media

I.  Getting started with analyzing social media networks



1.4  Individual contributions generate public wealth and risks

tools have been used successfully to create large-scale collaborative public projects like Wikipedia, open source software used by millions, new forms of political participation, and scientific collaboratories that accelerate research. Unheard of just a few years ago, today systems such as WeChat, Instagram, Facebook, wikis, Twitter and others are now headline news with social and political implications that stretch around the globe. Most use of these tools has been beneficial but costly exceptions do exist. Violent crowds focused on mob justice have been inflamed in many regions through the use of social media. Like film, radio, and TV before it, Internet social media can have positive and negative impacts. As social media matures, our task is to domesticate it, mitigating its worst costs while amplifying its value. Despite the very different shapes, sizes, and goals of the institutions involved in social media, the common structure that unifies all social media spaces is a social network. All of these systems create connections that leave traces and collectively create networks that can be visualized, analyzed and compared.

1.3  The rise of enterprise social media applications Social media are most visible in the form of consumer applications such as WeChat, Instagram, Facebook and Twitter, but significant use of social media tools takes place behind the firewalls that surround most corporations, institutions, and organizations. Inside these enterprises, employees share documents, post messages and engage in extensive discussions, annotate documents, and create extensive patterns of connections with other employees and other resources. Networked communication has become an indispensable link to customers and partners and a critical internal nervous system required for every aspect of commerce. Enterprise social media tools cultivate the internal discussions that improve quality, lower costs, and enable the creation of customer and partner communities that offer new opportunities for coordination, marketing, advertising, and customer support. As enterprises adopt tools like email, text and message boards, blogs, wikis, document sharing, and chat streams, they generate a number of social network data structures. Enterprise network datasets contain information about connections that can have significant business value by highlighting employees who play critical and unique roles. Some employees act as bridges or brokers between otherwise separated segments of a company. Others have patterns of connection that indicate that they serve as sources of information for many others. Social network analysis of organizations (often called Organizational Network Analysis or “ONA”) offers a

5

form of MRI or X-ray image of the organizational structure of the company (e.g., see Chapter 9). These images can illuminate the ways members of an organization are actually connected in contrast to the formal hierarchies of traditional “org-charts.” Technology consulting firms have highlighted the value of analyzing patterns of connection within an organization. The Gartner Group reported that social ­network analysis would prove to be a strategic advantage for a corporation, calling it an “untapped information asset.”1 They recommend the analysis of “business intelligence on the ties, information flows and value exchanges” within a corporation. Network analysis can be focused, they argue, on three separate regions of commerce: organizational network analysis, value network analysis, and influence analysis, which map loosely to internal, vendor, and consumer populations. In each segment, network analysis is a useful method for identifying choke points and positions of leverage, locating expertise, and enhancing innovation.

1.4  Individual contributions generate public wealth and risks Social media collective goods are a remarkable story of bottom-up individual initiative that leads to the creation of public value and wealth. Collections of individual social media contributors can create vast, often beneficial, yet complex social institutions. The intriguing challenge for the authors of this book and for a growing circle of social media analysts is to focus on individual behaviors while recognizing the emergent, collective properties of social media contributions. Seeing the social media forest, and not just the trees, branches, and leaves, requires tools that can assemble, organize, and present an integrated view of large volumes of records of interactions. Building a better view of the connected social media landscape can lead to improved user interfaces and policies that increase individual contributions and their quality. It can lead to better management tools and strategies that help individuals, organizations, and governments to more effectively apply social media to their priorities. And given increased awareness of intentional abuse and collective manipulation of these systems, often with political consequence, situational awareness of social media becomes a critical element for the hygiene of democratic discourse. Many utopian commentators have reported and proclaimed the benefits of social media. However, ­dangerous criminals, malicious vandals, promoters of racial hatred, and oppressive governments can also use social media tools to enable destructive activities. Critics of social 1 www.gartner.com, Using Social Network Analysis to Inform a Pattern-Based Strategy.

I.  Getting started with analyzing social media networks

6

1.  Introduction to social media and social networks

media warn of the dangers of lost responsibility and respect for creative contributions, when vital resources are assembled from many small pieces [1]. More recently, concern has focused on the ways malicious individuals, groups, organizations and even nation states can use social media to create impressions, shift understandings or erode trust [2]. While propaganda is an ancient human practice, new social media platforms change the speed, scale, scope, and possibly the effectiveness of adversarial information [3]. These dangers heighten the interest in understanding how social media phenomena can be studied, improved, and protected. Why do some groups of people succeed in using these tools while many others fail? How can successful groups resist invasion and internal division? Community managers and participants can learn to use social network maps of their social ­media spaces to cultivate their best features and limit negative outcomes. Social network measures and maps can be used to gain insights into collective activity and guide optimization of their productive capacity while limiting the destructive forces that plague most efforts at computer-mediated communications. People interested in cultivating these communities can measure and map social media activity in order to compare and contrast social media efforts to one another, and to learn from the best examples. Around the world, community stakeholders, managers, leaders, and members have found that they can all benefit from learning how to apply social network analysis methods to study, track, and compare the dynamics of their communities and the influence of individual contributions. Business leaders and analysts can study enterprise social networks to improve the performance of organizations by identifying key contributors, locating gaps or disconnections across the organization, and discovering important documents and other digital objects [4]. Marketing and service directors can use social media network analysis to guide the promotion of their products and services, track compliments and complaints, and respond to priority customer requests. Community managers can apply these techniques to public-facing systems that gather people around a common interest and ensure that socially productive relationships are established. Social media tools have become central to national priorities requiring government agency leaders to become skillful in building and managing their communities and connections. Governments at all levels must learn to optimize and sustain social media tools for public health information dissemination, disaster response, energy conservation, environmental protection, community safety, and more. The disturbing reality that many disaster events attract misinformation that can have deadly consequence motivates many efforts to better analyze the flow of social media to rapidly identify and deflect inaccurate and dangerous information [5].

In this book we explore social structure and organization through the application of the methods and concepts of social network analysis. Network analysis is a relatively recent scientific method for describing and analyzing a web of links among entities, including people. Network analysis provides powerful ways to summarize networks and identify key people or other objects that occupy strategic locations and positions within a matrix of links. Network visualizations can graphically map these structures to complement numerical measures and enable people to gain valuable intuitions and insights into the shape, size, density, sub-regions, and key locations within a connected population. Over decades, scientists, technologists, and entrepreneurs have dramatically improved the tools, analytic methods, visualization approaches, and sources of data for social network analysis. Increasingly potent software applications are available to study these phenomena and uncover useful, actionable insights. Until recently, these tools demanded significant programming and data management skills that excluded many interested users. We focus on a social network analysis tool designed for ease of use and customized for application to social media, the Network Overview Discovery and Exploration add-in for Excel (“NodeXL”). NodeXL embeds the tools and features needed to collect, analyze, visualize and report on a network within the familiar framework of the Excel spreadsheet.

1.5  Who should read this book Practitioners, researchers, and students interested in the study of social media can benefit from this book. Using this volume, business leaders, instructors, and students can apply principles of social network analysis to measure, analyze, and interpret real-world data from a variety of social media platforms. Readers will learn how to extract insights from networks to reveal internal business activity, external customer communities, and their local competitive landscape. Professors and instructors in a range of disciplines may find this volume useful in semester-long courses as well as shorter units related to computing, business, and social sciences. Technical classes in computer science/engineering, information science, human computer interaction, information visualization, cybersecurity, and even social physics have been increasingly focused on the topic of “social computing” and “computational social science”. In business and management schools, enterprise collaboration and customer communities remain important topics for generating business value. Digital humanities scholars and instructors have increasingly used social media to understand and enable novel connections

I.  Getting started with analyzing social media networks



1.6  Applying social media to national priorities

between people and cultural artifacts. In the social sciences, sociologists, anthropologists, historians, economists, demographers, political scientists, and other students of collective intelligence, collective action, and communities of practice have an opportunity to do data analysis at a scale never before possible with a less steep learning curve than demanded by programming languages.

1.6  Applying social media to national priorities Government agencies around the world are attracted to the possibilities of improved delivery of services at reduced costs but challenged by the loss of content control, liability for libel, pornography, or terrorist use. Open access to vast stores of government data expands their value while potentially calling attention to unfavorable information that could be used by political adversaries. Government professionals are excited by the prospects of increased citizen engagement but concerned by what that engagement may mean for their control over the flow of information and their obligations to protect privacy, avoid censorship while preventing libel, and other inappropriate uses of government information technology resources. Across the planet, developed and developing countries are attracted to the potential to use social media to change their societies, from promoting energy conservation and smoking cessation to new levels of political engagement with citizens. But as governments increasingly erode their protections for civil liberties and information security, and some prominent politicians effectively leverage social media in ways that increase division and amplify violent conflict, the positive applications of social media are balanced by challenges and a growing number of negative examples. Efforts by innovative citizen/residents have encouraged the idea that existing social media platforms can be harnessed for national priorities, such as disaster response. Social media has played a key role in major disasters ranging from Hurricane Katrina in New Orleans in August 2005, to the Tōhoku, Japan earthquake in 2011, to the California wildfires of 2018. Volunteers created websites to coordinate assistance, offer food, provide housing, share photos, create maps, help find transportation, and eventually help rebuild. Despite improved outcomes, modest design changes in the way information can be gathered, aggregated, validated, and shared could have increased the usefulness of social media. Disaster response is typically seen as a national government responsibility, but other services, such as community safety, are seen as a local government responsibility. Here too innovative individuals have ­created websites

7

and services to enable resident-to-­ resident assistance, such as Nextdoor,2 which describes its effort as follows: Nextdoor is the best way to stay informed about what’s going on in your neighborhood—whether it’s finding a last-­ minute babysitter, planning a local event, or sharing safety tips. There are so many ways our neighbors can help us, we just need an easier way to connect with them.

Another successful community safety effort has been Amber Alert,3 named after a child who was abducted and murdered in 1996. This alerting system, now coordinated by the U.S. Department of Justice, claims to have directly assisted in almost 500 safe returns of abducted children. It may also have raised awareness enough to have prevented many other abductions. Resident reports are being solicited for tornadoes, earthquakes, floods, or other natural disasters as well as for reporting on fraud, abuse, and waste of government funds. Positive contributions such as fixes to the Library of Congress Card Catalog or the National Library of Medicine’s PubMed service lead the way in suggesting further possibilities. The innovative Peer-to-Patent system (see Noveck's 2009 book on WikiGovernment [6]) invited specialists in certain technical areas to contribute information on prior art related to patent applications. Then a group discussion ranked the 10 items for submission to the patent examiners so as to speed up and improve their work at the U.S. Patent and Trademark Office. Noveck summarizes her case this way: “Ordinary citizens have more to offer than voting or talking. They can contribute their expertise and, in so doing, realize the opportunity now to be powerful… Collaborative governance is an idea whose time has come.” [6, p. 190]. Volunteers to museums, parks, hospitals, or schools could also improve public services at national and local sites. Government-run websites such as http:// nationalservice.gov now facilitate these service efforts. Data shows that a large percent of residents volunteer in the most active states (e.g., 43% in Utah; 35% in Minnesota, Wisconsin, and South Dakota), while the least active states lag far behind (e.g., Nevada, New York, Florida, and Louisiana all under 20%). Could increased visibility or awareness of volunteer efforts increase participation?4 Millions citizen scientists donate their resources, time, and skills to help classify galaxies, identify exoplanets, map invasive species, transcribe historical records, map neurons, and identify a cure for cancer. 2 https://nextdoor.com. 3 www.amberalert.gov, America's Missing: Broadcast Emergency Response. 4 https://www.nationalservice.gov/vcla/state-rankings-volunteer-rate.

I.  Getting started with analyzing social media networks

8

1.  Introduction to social media and social networks

Sites like Zooniverse5 have attracted over 1.7 million community members have helped classify over 400 million items through their growing collection of projects. For nearly 2 decades, the SETI@home project has allowed volunteers to donate their computer's resources to analyze radio telescope data that aids in the search for extraterrestrial intelligence.6 The Smithsonian Institution has created the ambitious Encyclopedia of Life7 project to achieve Edward O. Wilson’s goal of a web page for each of the estimated 1.8 million species on earth. The payoffs in scientific knowledge to support biodiversity and environmental preservation are potentially large, but the challenges of getting professional scientists, citizen scientists, and nature enthusiasts to work together are also substantial. Designers and community managers continue to search for the motivational structures and recognition strategies needed to gain broader participation. Social networks also support rapid dissemination of public information (e.g., on flu vaccinations, weather alerts, or community safety threats). Many people are more likely to trust and act on cell phone calls or email messages from friends and family than from pronouncements by public officials on television programs or newspaper reports. Public officials can also disseminate less time-sensitive information on energy conservation strategies, environmental protection initiatives, or health alerts about obesity reduction or smoking cessation. Citizen-generated YouTube videos often have more impact than carefully scripted appeals by professionals at press conferences. Analysis of view counts and comments can show how effective various strategies are in reaching different demographic segments. Leaders of many non-governmental community groups have come to appreciate the growing power of social media with their increasingly rich services. Communities can become energized by modern technology-mediated versions of parent-teacher as­ sociations, neighborhood watches, and disaster planning teams. Even smaller groups such as book clubs, high-school orchestras, or local birdwatchers benefit from use of communications tools such as Twitter feeds, Facebook pages, or Google groups.

1.7  Worldwide efforts Although this book emphasizes examples that we are more familiar with in the United States, there are worldwide efforts to apply social media strategies and encourage further research and development. About a

decade ago, during a time of relative optimism about the potential of the disruptive social media technologies, the European-based Institute for Prospective Technological Studies produced two thoughtful reports on The Impact of Social Computing on the EU Information Society and Economy8 and Public Services 2.0: The Impact of Social Computing on Public Services.9 The first report suggested that “social computing could play an increasingly important role in re-engaging citizens in political debate, in securing social cohesion and harmony, and it could provide a platform for dialogue on the grand challenges of the EU [European Union] and the rest of the world.” The second report encouraged “policy makers to seize the opportunities of social computing but also to mitigate any undesirable effects” and laments the “limited provision of citizen-centered public services by governments.” The report further warns “that the empowerment and transparency characteristics of social computing initiatives seem to disrupt existing power balances.” In the coming years, the impact of social media on political systems became apparent through the Arab Spring and the weaponization of social media by ISIS. Meanwhile, a series of leaked documents by Edward Snowden and others revealed the extent to which the United States and other nationstates were collecting massive amounts of social media data. Additional cybersecurity breaches of social media websites and the clear abuses of privacy data by key social media companies including Facebook, helped pave the way for the EU's General Data Protection Regulation (GDPR) legislation that became effective in 2018. Hailed as the most comprehensive data privacy legislation in the past 20 years, the new law has far reaching implications on the ways in which companies collect and use social media data. Research efforts led by groups such as the European Society of Socially Embedded Technologies (www. eusset.eu) and international Web Science Trust (www. webscience.org) have helped shift the attention of researchers and policy makers from a purely technical focus to a social and socio-technical focus. Early position papers [7, 8] emphasized the need to develop strong scientific foundations for social media research that integrates sociotechnical systems thinking; a call that has largely been answered through researchers affiliated with the CSCW and ECSCW research communities. Hendler et al.’s [9] emphasis on the social nature of web technologies has been proven true again and again in the past decade: The social model enabled by humans interacting in ways allowed by that technology is more difficult to explain the success or failure of the sites hinges on the rules, policies, and user

5 https://www.zooniverse.org/. 6 https://setiathome.berkeley.edu/.

8 http://ftp.jrc.es/EURdoc/JRC54327.pdf.

7 www.eol.org.

9 http://ftp.jrc.es/EURdoc/JRC54203.pdf.

I.  Getting started with analyzing social media networks



References

c­ommunities they support. Given that the success or failure of Web technologies often seems to rely on these social features, the ability to engineer successful applications requires a better understanding of the features and functions of the social aspects of the systems.

The authors of this book see many opportunities for diverse academics and professionals to contribute to our understanding of how social media are already changing our world. We also see an active role for them in the design of future technologies, as well as future social, economic, and political systems. Methods such as social network analysis applied to social media datasets will no doubt be a key contributor to such efforts. Technology promoters succeed more often when they address usability and sociability, are alert to human needs and values, and are sensitive to balancing policies and norms.

1.8  Practitioner’s summary Existing social institutions, educational curricula, business plans, and government policies are shifting as a result of social media tools and their application. Forward-looking universities are adding courses on the study of social computing, social informatics, new media, and digital society. Journal editors, conference organizers, and national science funding agencies are working to take advantage of the opportunities of using these new tools and techniques. Individuals, organizations, and government agencies are devoting resources to using social media for their benefit while avoiding the dangers. Understanding how these social media networks thrive, change, or fail is a substantial challenge to researchers and professionals. Researchers in social network analysis have provided a set of concepts and metrics to systematically study these dynamic processes. Innovators in information visualization have also contributed to helping users to discover patterns, trends, clusters, gaps, and outliers, even in complex social networks. Each day solutions for better network insights are being found that bring competitive advantages to business product developers, opportunities for government agency staffers, and new possibilities for nongovernmental social entrepreneurs.

1.9  Researcher’s agenda Now is an exciting time for those involved in the emerging discipline of social media network analysis. Researchers are designing novel collaborative technologies and social strategies that enable new forms of working and playing. They are also analyzing existing communities to find out what strategies and design decisions lead to success and avoid social problems.

9

Although social media success stories abound, there are countless examples of failed attempts to effectively apply social media to achieve desired goals. Social network analysis offers a systematic method to evaluate social media efforts, replacing anecdotes with scientifically based evidence. Unfortunately, many observers see no urgency in changing business strategies, marketing plans, research directions, curricula, or government programs. We hope this book will change their minds by showing the compelling business opportunities, attractive research challenges, strong educational needs, and important national priorities that social media can address and network analysis can elucidate. Just as the physicists of the 1940s were challenged and troubled by the awesome forces they unleashed, researchers studying these social phenomena may yet create technologies that release human chain reactions, harnessing vast amounts of human energy to overcome the social problems that challenge our world. In the past 400 years, scientists have focused on fundamental physical phenomena, such as gravity, magnetism, nuclear forces, and genetic information. Their work has produced profound changes in human life as we gain insights and control over core physical forces. Cellular communication networks, material sciences, and nuclear power are leading examples of the accomplishments of this vast intellectual endeavor. Similarly, biologists have revealed the core processes of all life, exposing the structure of DNA and opening the door to powerful techniques and practices that are only just unfolding. Because powerful technologies are eagerly sought by those who might put them to destructive purposes, urgent efforts are needed to ensure constructive outcomes. We believe that open discussion, broad participation, and respect for individual rights can help lead the way to more beneficial results.

References [1] J.  Lanier, You Are Not a Gadget: A Manifesto, Knopf Publishers, New York, 2010. [2] P.W.  Singer, E.T.  Brooking, Likewar: The Weaponization of Social Media, Eamon Dolan/Houghton Mifflin Harcourt, New York, NY, 2018. [3] Y.  Benkler, R.  Faris, H.  Roberts, Networked Propaganda: Manipulation, Disinformation, and Radicalization in American Politics, Oxford University Press, Oxford, 2018. [4] R.S.  Burt, Structural holes and good ideas, Am. J. Sociol. 110 (2) (2004) 349–399. [5] K.  Starbird, J.  Maddock, M.  Orand, P.  Achterman, R.M.  Mason, Rumors, False Flags, and Digital Vigilantes: Misinformation on Twitter After the 2013 Boston Marathon Bombing, in: iConference 2014 Proceedings, 2014, pp. 654–662. [6] B.  Noveck, Wiki Government: How Technology Can Make Government Better, Democracy Stronger, and Citizens More Powerful, Brookings Institution Press, Washington, DC, 2009. [7] T. Berners-Lee, W.T. Hall, J.W. Hendler, N. Shadbolt, D. Weitzner, Creating a science of the web, Science 313 (5788) (2006) 769–771.

I.  Getting started with analyzing social media networks

10

1.  Introduction to social media and social networks

[8] N. Shadbolt, T. Berners-Lee, Web science emerges, Sci. Am. (2008) 32–37. [9] J.W. Hendler, N. Shadbolt, W.T. Hall, T. Berners-Lee, D. Weitzner, Web science: an interdisciplinary approach to understanding the world wide web, Commun. ACM 51 (7) (2008).

Additional resources Y.  Benkler, The Wealth of Networks: How Social Production Transforms Markets and Freedom, Yale University Press, New Haven, CT, 2005. M. Castells, The Rise of the Network Society, Blackwell, Malden, MA, 1996. N.  Christakis, J.  Fowler, Connected: The Surprising Power of Our Social Networks and How They Shape Our Lives, Little, Brown, New York, NY, 2009. R.L. Cross, A. Parker, R. Cross, The Hidden Power of Social Networks: Understanding How Work Really Gets Done in Organizations, Harvard Business Press, Boston, MA, 2004. J.  Kleinberg, The convergence of social and technological networks, Commun. ACM 51 (11) (2008) 66–72.

J.  Preece, B.  Shneiderman, The reader-to-leader framework: motivating technology-mediated social participation, AIS Trans. Human Comput. Interaction 1 (1) (2009) 13–32. Available at http://aisel. aisnet.org/thci/vol1/iss1/5. R.D.  Putnam, Bowling Alone: Collapse and Revival of the American Community, Simon and Schuster, New York, 2000. L. Rainie, B. Wellman, Networked: The New Social Operating System, The MIT Press, Cambridge, MA, 2014. B.  Shneiderman, C.  Plaisant, M.  Cohen, S.  Jacobs, N.  Elmvquist, N.  Diakopoulos, Designing the User Interface: Strategies for Effective Human-Computer Interaction, sixth ed., Addison-Wesley, Boston, MA, 2016. J.  Surowiecki, The Wisdom of Crowds, Anchor Books, New York, 2004. S. Turkle, Alone Together: Why We Expect More from Technology and Less from Each Other, Expanded, revised ed., Basic Books, New York, NY, 2017. L.  Palen, A.L.  Hughes, Social media in disaster communication, in: Handbook of Disaster Research, Springer, Cham, 2018, pp. 497–518.

I.  Getting started with analyzing social media networks

C H A P T E R

2 Social media: New technologies of collaboration O U T L I N E 2.1 Introduction

11

2.2 Social media defined

12

2.3 Social media design framework 2.3.1 Size of producer and consumer population 2.3.2 Pace of interaction 2.3.3 Genre of basic elements 2.3.4 Control of basic elements 2.3.5 Types of connections 2.3.6 Retention of content

12

2.4 Social media examples 2.4.1 Asynchronous threaded conversation 2.4.2 Synchronous conversation

18 18 20

13 14 14 15 16 17

2.1 Introduction Technologies that support social interaction are one of the marvels of our time. The unprecedented development and use of social mediating technologies have engendered radically new ways of working, playing, and creating meaning, leaving an indelible mark on nearly every domain imaginable. Billions of people now weave a complex collection of email, Instagram, Facebook, Twitter, WeChat, WhatsApp, mobile short text messages, shared photos, podcasts, audio and video streams, blogs, wikis, discussion groups, and virtual reality game environments to connect them to the world and the people they care about. Increasingly, people access these tools using mobile devices that can tie content to locations in real time. Behind organizational firewalls, a host of enterprise social media tools echo the social media tools so popularly used in the public Internet. The novel ways that people have adopted and adapted these technologies to their particular needs is a testament to human ingenuity and sociability. Despite the growing ubiquity of social technologies, their potential has still hardly been tapped. Effectively

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00002-9

2.4.3 2.4.4 2.4.5 2.4.6 2.4.7 2.4.8 2.4.9 2.4.10 2.4.11

The World Wide Web Collaborative authoring Blogs and podcasts Social sharing Social networking services Online markets and production Idea generation Games and virtual worlds Mobile services

21 22 22 23 24 25 26 26 27

2.5 Practitioner’s summary

27

2.6 Researcher’s agenda

28

References

28

Additional resources

29

using and improving social technologies is far from trivial. A complex interplay between social practices and technological infrastructures takes place within each of these platforms. Architects will tell you that the physical design of a building or city can dramatically influence the ways in which people interact with one another. Teaching a course in a room with seats arranged in a circle vs seats arranged in rows facing the front invites a different form of participation from students. Although the physical layout does not wholly determine the forms of interaction, it does make certain interactions easier and others more challenging. Similarly, the sociotechnical infrastructure, or platform, that underlies online activity influences social interaction. This is not an argument for technological determinism. Rather, it is a solid materialism that recognizes that technologies change the fabric of the material world, which in turn changes the social world. For example, microblogging sites like Twitter enable short exchanges ideal for efficiently pointing out resources or knowing what events other people are attending, while discouraging in-depth discussion and analysis on the

11

© 2020 Elsevier Inc. All rights reserved.

12

2.  Social media: New technologies of collaboration

platform itself. In contrast, traditional blogs without length limitations and with their support for sharing multimedia content and comments are better suited for more in-depth presentations and conversations. Other media including books, newspapers, wikis, email, social networking sites, and so forth each have a set of properties that create a unique terrain of interaction. Learning to effectively meet your objectives using social media requires an understanding of that terrain and the social practices that have grown up around its use. One of the most exciting aspects of online social media tools is that they produce an enormous amount of social data that can be used to better understand the people, organizations, and communities that inhabit them. More specifically, they create relational data: information about who knows or is friends with whom, who talks to whom, who hangs out in the same places, and who enjoys the same things. These relational datasets provide a wealth of new opportunities to understand and improve the social worlds we inhabit, as discussed throughout this book. The purpose of this chapter is to introduce some of the important social media systems and provide a language and framework to talk about their key properties. It is also intended to informally introduce the relationship between social media systems and the networks they implicitly and explicitly create. This chapter begins with a definition of social media, followed by a framework for characterizing types of social media, and then takes you on a whirlwind tour of several important social media technologies that have emerged in the recent past.

2.2  Social media defined Social media refers to a set of computer-network based tools that support social interaction between users. The term is often used to contrast with more traditional media such as television and books that deliver content to mass populations but do not facilitate the creation or sharing of content by users. Social media is about transforming broadcast (one-to-many) into dialog (many-to-many). In practice, “social media” is a catchall phrase intended to describe the many online sociotechnical systems that have emerged in recent years, including services like email, discussion forums, blogs, microblogs, texting, chat, social networking sites, wikis, photo and video sharing sites, review sites, and multiplayer games. Other terms are also used to describe many of these systems including “Web 2.0,” the “read/write web,” “social computing,” “social software,” “collective action tools,” “sociotechnical systems,” “computer-mediated communication,” “groupware,” “computer supported cooperative work (CSCW) systems,” “virtual” or “online communities,” “usergenerated content,” and “consumer-generated media.”

Pioneers of the information age such as Vannevar Bush who envisioned a hypertext-like device called the “memex” [1] and Douglas Engelbart who saw a future of graphical interfaces (i.e., windows), computer mice, and multipleauthored digital content [2] decades before it was realized, were interested in augmenting human intellect. In other words, they wanted to develop systems that “increase the capacity of man to approach a complex problem situation, to gain comprehension to suit his particular needs, and to derive solutions to problems” [2]. These goals have slowly been realized through remarkable developments in hypertext, human-computer interaction, the World Wide Web, and mobile technologies [3]. As the world has become increasingly connected, the focus has shifted from a focus on one person interacting with a computing machine to augmenting social interaction, experience and collective intelligence. Social media tools enable users to collaboratively create, find, share, evaluate, and try to make sense of the mass of information available online. They also allow users to connect, inform, inspire, and track other people and topics. The new blend of social action and technological infrastructure allows entirely new ways of collaborating. Users can receive personalized recommendations based on the prior purchasing habits of thousands of other “similar” people, identify high-interest news stories based on real-time voting by the crowd, collaboratively author the world's largest and most-read encyclopedia, and instantly notify hundreds of followers about an online video presentation they found insightful. Unfortunately, people and governments can also more effectively deceive others, radicalize individuals, and bully victims. Our hope is that as we augment social interaction, we will promote positive interactions, while minimizing negative ones.

2.3  Social media design framework Social media systems come in a variety of forms and support numerous genres of interaction. Although they all connect individuals, they do so in dramatically different ways depending, in part, on the technical design choices that determine questions like these: Who can see what? Who can reply to whom? How long is content visible? What can link to what? Who can link to whom? As discussed in the introduction, these design choices can influence the social interactions that they enable and mediate. In addition, social practices, personalities, and history heavily influence how social media systems are used. If designers have learned anything from successful social media systems like email and discussion forums, it is that they can be adapted to meet a surprisingly wide array of individual and community needs. Despite the adaptability of many social media systems, it is important to distinguish among systems as different as email,

I.  Getting started with analyzing social media networks



wikis, and massively multiplayer video games while recognizing their similarities. One way to make sense of the bewildering proliferation of systems and services is to consider a set of key dimensions along which many social media services can be located. This approach provides a language and framework for comparing social media tools. This section considers six key dimensions: • • • • • •

13

2.3  Social media design framework

Size of producer and consumer population Pace of interaction Genre of basic elements Control of basic elements Types of connections Retention of content

These are not the only dimensions of possible interest, but they capture many of the important differences between social media tools. They also help lay the groundwork for the remainder of the book, which will use more formal methods to analyze the networks that are implicitly or explicitly created by the various social media platforms.

2.3.1  Size of producer and consumer population In most social media systems, producers and consumers are drawn from the same set of users. Users are producers one moment and consumers the next. However, differentiating between those who produce and consume content can be useful in comparing social media systems, even if the set of producers and consumers are not mutually exclusive. Social media services vary in terms of their intended number of producers and consumers. An email is usually

authored by just one person, whereas a wiki document is likely to be authored by several or even hundreds of people. An individually authored email might be sent to just one other person or be broadcasted to thousands. More generally, social media tools support different scales of production and consumption of digital objects. Table 2.1 provides some examples of social media systems, as well as some traditional media systems, and where various actions related to them fall within the producer and consumer size dimensions. You may notice that some systems show up in different places based on their usage scenario or the features that are being discussed. Many social media tools help individuals or small groups interact. Text messaging, video chat, and personal or “direct” messaging within general-purpose social networking sites provide intimate communication channels comparable to phone calls and face-to-face office meetings. Social media can help individuals reach out to medium-sized groups of friends or acquaintances by broadcasting a personal message (e.g., a tweet sent to a user’s followers on Twitter; a post sent to a departmental email list; an instagram photo shared with family and friends) or allowing others to overhear a comment (like a post to someone’s Facebook feed). They can also allow individuals to potentially reach large groups through popular blog posts, podcasts, videos posted on sites like YouTube, or updates on Twitter by companies or celebrities with numerous followers. Other social media tools help medium-sized groups reach out. Pages on social networking sites like Facebook and Twitter allow a group (i.e., your friends or those you follow) to create a personalized stream of information customized for you. Other tools like online surveys help

TABLE 2.1  Examples of social media and pre-digital media systems organized by the size of producer and consumer populations Size of consumer population

Size of producer population Small

Medium

Large

Small

Instant messaging Personal messaging (e.g., within Facebook) Video conferencing Phone call Face-to-face office meeting

Committee report to a decision maker Online survey Social networking friend feed Twitter follow feed

Professional services reports for decision makers Personalized suggestions based on recommender systems

Medium

“Social” or family blog Stack Overflow Q&A Departmental email list Tweet sent to followers Facebook post Twitch stream

Group blog on niche topic Internet relay chat room Internal department wiki Facebook group Niche YouTube channels Local markets (e.g., Craigslist)

Professional report for specialty group Zooniverse citizen science project Idea-generation sites (e.g., IdeaConnection)

Large

Popular blog or podcast Message to large forum or email list Popular Twitter user’s tweet Popular YouTube video Company website Novel or newspaper

News rating site (e.g., Reddit) Wikipedia page Television program Popular discussion forum User-generated databases (e.g., IMDB) or marketplace (e.g., Threadless)

Large online marketplace (e.g., eBay) Wikipedia YouTube FamilySearch Indexing Popular massively multiplayer game

I.  Getting started with analyzing social media networks

14

2.  Social media: New technologies of collaboration

aggregate information from many people for a small group of people or an individual analyzing the data. Group blogs or collections of related blogs about niche topics within the blogosphere allow a medium-sized number of bloggers and commenters to interact with one another. A number of different tools facilitate interaction between medium-sized groups whether they are part of a Facebook group, YouTube channel, Slack or Internet Relay Chat (IRC) room, or combinations of tools as in an enterprise social networking site. A department or workgroup wiki can allow members or co-workers to coauthor materials that are of interest to their group. Finally, some tools enable medium-sized producer groups to reach large consumer groups in a way similar to TV programs that take considerable effort to produce, but can reach large audiences. Some of these include online databases such as the Internet Movie Database (IMDB), where user-generated movie content is shared with the world; news sites like Reddit, where dozens to hundreds of people recommend a given news article that is consumed by much larger populations of read-only users; discussion forums, where posts by a few dozen active members may be seen by thousands of readers; and Wikipedia pages that are edited by dozens of people and read by thousands. Some of the most interesting social media tools are those that help harness the power of the masses. For example, some recommender systems (e.g., Netflix, Spotify, Amazon, MovieLens) provide personalized suggestions of books, movies, or songs by comparing your ratings with ratings’ of other users. Other large groups help generate ideas that are used by medium-sized groups such as small businesses, corporate departments, or government agencies. For example, Kaggle is a website that allows anyone to start a sophisticated data analysis project with a reward so that many motivated data analysts can contribute ideas to solve their problems. NASA’s ClickWorkers help identify craters on Mars, Zooniverse contributors categorize the shapes of astronomical observations, and iNaturalist users report animal sightings. Many of the most well-known social media sites allow large producer groups to interact with large consumer groups. Although there are many more Wikipedia readers than contributors, both groups are large. Online marketplaces like eBay allow the masses to sell and purchase goods. Meanwhile, social sharing sites like YouTube, Instagram and Flickr make it easy for large numbers of producers and even larger numbers of consumers to interact. While these sites often facilitate small group interaction, they also aggregate those interactions so that you can search and navigate through large corpora of usergenerated content. Massively multiplayer games rely on having large numbers to produce content and social experiences that make use of the entertaining environment.

2.3.2  Pace of interaction The pace at which interaction occurs is another important dimension along which researchers organize social media systems. Traditionally, researchers distinguished between asynchronous and synchronous communication. Asynchronous systems like email, discussion forums, and voicemail presume a staccato pattern of interaction spread out over hours or days or weeks. Though less immediate, these systems have the advantage of allowing you to schedule your participation without much coordination with other people who may be in a wide range of time zones. They also potentially encourage more careful contributions. In contrast, synchronous systems, like chat, instant messaging, videoconferencing, multiplayer games and graphical worlds, require that partners interact at the same time, as in face-to-face interactions and telephone calls. Although they require temporal coordination, they can create a richer environment for interaction as participants quickly react and adjust to one another’s signals in near real time. The pace of interaction has implications for the kinds of groups that form using each kind of tool. Global collaborations are often easier using asynchronous tools that don’t require people to change their sleeping habits. But some interactions need more rapid turn-taking to accomplish their goals. More recently, the distinction has become increasingly blurred. For example, Twitter users often reply within minutes to another’s tweet, but it is completely acceptable to reply a day later as well. Replying to a Facebook post or status update on other social networking sites is similar in this regard. Tools like Google Hangouts are now integrated with the widely used Gmail web email system, again blurring the distinction between synchronous and asynchronous modes of communication by making it easy to move from one mode to the other. Apps like Marco Polo facilitate the sharing of video messages, which can also be viewed live. However, users’ varied expectations about the pace of interaction within these tools remains important for understanding social media environments.

2.3.3  Genre of basic elements Digital objects, the basic elements of social media systems, vary in size and type. Twitter posts (i.e., tweets) are limited to 280 (initially 140) characters, whereas email messages are typically a few lines to a few paragraphs in length but can be even longer. This difference in size produces dramatically different patterns of interaction. Instant messaging design choices such as the size of the text box and messaging window promote brevity. Short messages are often directed to other people who are assumed to be busy and engaged in other activities. Meanwhile, MediaWiki (the wiki platform used by

I.  Getting started with analyzing social media networks



2.3  Social media design framework

Wikipedia) supports six levels of headers and automatically generates a table of contents, making it relatively easy for you to create large pages to accommodate complex documents and concepts. Using the digital objects supported by social media tools is another way for you to learn about the similarities and differences among them. Social media systems have often evolved around a distinct type of digital object: short posts or text messages from people and companies on Facebook and (even shorter) Twitter, videos at YouTube, photos at Instagram and Flickr, bookmarks (i.e., website URLs) at Pinterest, books at Amazon, music or podcasts at iTunes, TV shows at Hulu, messages in discussion forums or email lists, pages at Wikipedia, products at eBay, presentations at SlideShare, 3D objects in Second Life, and career professionals at LinkedIn. Over time, each has expanded into additional modes of interaction and types of digital objects. Instagram and flickr were once just for still images and both accept videos now. Business sites expand into lifestyle and leisure areas. Over time many functions have been rolled into the major platforms, marginalizing standalone platforms that specialize in just one mode of interaction or type of data. Each of these platforms provides you with different levels and mechanisms of engagement and interaction. For example, virtual worlds more closely model embodied physical interactions, where avatars can convey meaning through proximity and orientation [4]. They also introduce many of the burdens of face-toface interaction, demanding attention to successfully puppet the avatar in interactions with other partners [5]. Although these differences may relate to the type of media (e.g., video, audio, text, 3D model), there are further distinctions within each type. Wikis support structured text elements like tables and bullets, whereas email typically does not. Some virtual worlds intended for children use cartoon characters, while multiplayer games with more mature users like World of Warcraft include realistic-looking creatures. Of course some social media systems like Facebook include many basic elements: profile pages, wall posts, personal messages, applications, instant messages, notes, groups, photos, tags, status updates, and so on. Wikipedia has user pages, talk pages, articles, edits, categories, and so forth. Even in these systems, identifying the basic elements of the system is important because they give you the building blocks for your interactions. They are also the building blocks of networks when they are connected together or exchanged, as you will learn throughout this book.

2.3.4  Control of basic elements Social media systems provide different levels of control over their basic elements. They can restrict who can create, edit, read, invite, respond to, subscribe to, and

15

share content of various types. Some systems differentiate between anonymous users, registered users, and those with special privileges such as administrators. For example, some discussion forums require that users log in before they post, but they allow anyone to read the messages created by the community. This helps reduce spam by creating a higher barrier to entry, while still allowing anyone access to the content. It also allows you to exclude participants you define as violating your social norms and expectations. In discussion communities that focus on sensitive topics (e.g., patient support groups), you can limit access to content until a person is registered, a process that may also require some type of approval process by current administrators. Other systems like eBay require users to provide validated credit card information before they can sell items. The more open a community, the more potential there is for deviant behavior as evidenced by the frequent spam sent to wikis, email lists and early forms of online discussion spaces like Usenet. However, closing a community off too much may reduce the number of contributors, whereas openness may attract high-quality contributions that combat the effects of spam and abuse. This happens in the many high-value pages in Wikipedia, where poor edits left by non-registered users are quickly reverted by other registered and non-registered users. Choosing the right types of barriers to entry is an important part of online community building [6]. For example, in studying real-world communities, Ostrom [7] found that successful communities had clearly defined boundaries, largely to overcome problems associated with outsiders taking advantage of internally produced or maintained resources. Boundaries are also important in that they encourage frequent, ongoing interaction among group members. This is critical because repeated interaction is perhaps the single most important factor in encouraging cooperation [8]. If individuals are not likely to interact in the future, there is a huge temptation to behave selfishly and free ride. On the other hand, knowing that you will be interacting with others on a continual basis can lead you to create a reputation, which serves as a powerful deterrent to short-run, selfish behavior. Boundaries can have an impact on the kinds of interactions in which people are willing to engage because of the ways they shape the expected audience. Some media, like telephones and private postal mail, encourage the expectation that only a specific group of selected others will be the audience to your message. Other media, like tweets or Reddit posts, are likely to be seen by any number of unknown people. Further, some media prevent the identity of message creators to be known with certainty, if at all (e.g., pay phones or anonymous letters). Control structures can heavily influence governance and the distributed or centralized nature of the

I.  Getting started with analyzing social media networks

16

2.  Social media: New technologies of collaboration

environment. Although email lists, message boards, and Stack Overflow are all examples of asynchronous threaded conversation (see Section  2.4.1), their control structures are different. In centralized systems like email lists and many discussion forums, all communication flows through a single point, which is controllable by a single person (administrator) or small group who can wield dictatorial control over resources (messages) and access (who can subscribe). As a result, email list owners can remove people for inappropriate conduct or prevent spam or other inappropriate messages from being sent out or stored in the archive. They often serve as “benevolent dictators.” In contrast, distributed systems, like the Usenet network on the early text-only Internet, or the more current Blockchain platforms, are composed of hundreds of thousands of interlinked systems all interacting with a set of neighbors. Participants of sites like Stack Overflow use ratings to help identify high or low quality content. The lack of a central point of control makes it impossible or difficult for a member to exclude others or remove content. The granularity of control is another important factor. Users of a wiki can edit individual characters of a shared document, whereas other systems limit users to authoring entire messages that cannot then be edited. Twitter users can follow another user and then receive all tweets from that user. In many systems, users can only edit their own content, whereas in other systems, such as wikis and Google Documents, users can edit others’ messages, documents, or objects. The level of granularity may differ for different user groups within the same system. For example, discussion boards on many websites create a preset number of containers (referred variously as “folders,” “topics,” “groups”) for interaction or leave control over the creation of new spaces in the hands of a small number of administrators. In contrast, email lists can be configured to allow anyone to start new threads simply by sending a new message that is not a reply to a previous message. Of course, systems in

which users can create spaces with little restraint often contain many more of such spaces. The pace of interaction can be crossed with the granularity of user control to characterize systems (see Table 2.2).

2.3.5  Types of connections There are many ways that the basic elements of social media systems can be connected. It is important to understand these connections or “ties” in order to construct and understand networks from each kind of social media system. The next chapter goes deep into the theory and language of networks. This section describes the many types of connections that exist in social media systems and explores the ways collections of these connections create larger social systems that you can analyze with the math, tools, and insights of social network analysis. The basic elements of many social media systems can be connected to one another explicitly or implicitly. Users intentionally and knowingly create explicit connections, whereas implicit connections are inferred from the details of many digital traces. Perhaps the most common type of explicit social media connection is friending on social networking sites, where both people must approve the connection before it is realized. Other examples of explicit connections are following another user on Twitter, hyperlinking a wiki page to another page, tagging two photos or videos with the same tag, and adding someone to a text chat group. Implicit connections can be inferred when a user sends another user an email message, “favorites” content (and by extension its author), replies to a discussion post, or “Likes,” “Loves,” or “Upvotes” another user or their content as some sites allow. Although these actions are intentional, they are not performed with the explicit intention of creating a connection with the person. Other more subtle implicit connections can be identified, such as connecting people who “hang out” in the same discussion forums or Facebook groups, or who edit the same

TABLE 2.2  Examples of social media categorized by the pace of interaction and the granularity of control over content Pace of interaction

Granularity of control Fine

Medium

Coarse

Users can directly ­control smallest units of content ­(characters, pixels, bytes)

Users control medium-sized blocks of content (objects, attributes, tracks, players) that they can only indirectly alter or that can be altered by other users

Users control large block of content (documents, messages, blog posts, photos), rarely edited or modified by others

Synchronous

Real-time shared canvas

Virtual worlds, multiplayer games, real-time networked musical jamming

Chat, instant messaging, texting, Twitter

Asynchronous

Shared documents (e.g., Google Docs), source code, Wikipedia

Contribution to collected works like an album, anthology, report section, discussion group, or photosets

Email; blog posts and comments; sharing of links, photos, videos, and documents; turn-based games

I.  Getting started with analyzing social media networks



2.3  Social media design framework

wiki pages, or people can be connected by the books they both purchased from an online bookstore. These individuals may not know one another, but they are connected by their shared interests, locations and activities. Other connections can be inferred from data that are often not public but are available to the hosts or owners of social media systems such as reading patterns of discussion forums, music downloads, patterns of telephone calls, and location information. Smartphones with location sensors allow platforms to create implicit links between people who go to similar types of places, even if they do not go there at the same time. In innumerable ways, users now leave behind traces that form an intricate web connecting every person with the other people, locations, and digital objects around them. An important distinction among the types of connections people can create using social media platforms is between directed and undirected links. If you and another person become friends on Facebook, the connection is a mutual one. In other words, it is undirected. Likewise, if you both are tagged as an “expert,” then you are connected by an association that is mutual and thus undirected. In contrast, some systems like Twitter allow people to follow other users without first gaining those users’ approval. This creates a different type of tie, where the directionality of the tie is important (i.e., who is following whom). Directed ties are also created when a person invites another person, favorites content, and creates a hyperlink from one page pointing to another page. In all of these cases, connections flow from one person or object to another and may not necessarily be reciprocated. Finally, connections mean different things and can have different weights and values. For example, two people on Facebook can either be friends or not be friends; it is a binary connection that is either on or off. In contrast, two Facebook friends may send each other personal messages. The strength of their messaging connection could be measured based on the number of messages or the number of different days they each sent one another messages. These are examples of weighted connections that vary in intensity. These weights often contain important information about the strength of a tie. For example, if Marc sent 10 messages to Ben last week and only 1 to Derek and 3 to Itai, it is probably safe to say that last week Marc was more strongly connected to Ben than to Derek and Itai (at least via that messaging medium). The examples shared so far primarily connect people to each other, objects to each other, or people to objects. Recently, location has become an expanding part of social media services, allowing connections to be created between people, objects, and places. Smartphones are opening a new era of social media that integrates information about location and activity in novel and powerful ways. New kinds of ties are being formed by just

17

being in the same place as someone else, even at different times. Just having a phone or a laptop nearby someone else’s can create implicit connections.

2.3.6  Retention of content Social media systems also vary in how long content is retained. On one end of the spectrum are systems like wikis that typically create a permanent history of all actions that occurred in the system. Not only is each action recorded and stored, it is made available on article history pages and user contribution pages. At the other end of the spectrum, some instant messaging or voice-over Internet Protocol (IP) systems do not centrally record the interactions at all, allowing for fleeting exchanges more reminiscent of most face-to-face conversations. Users at the end points to these conversations can, of course, record them but extra effort must be taken. Many social media systems fall somewhere in the middle. For example, as of this writing, searches of the Twitter network via the public free API used by many software analysis tools can only receive from the most recent 8 or 9 days. The horizon of the past in these systems is in flux as data volumes grow along with information processing capacity and commercial requirements. The desire to add social media data to our long-term cultural memory has prompted interesting partnerships, as evidenced by the agreement between Twitter and the Library of Congress. However, these arrangements have not made widespread long-term historical social media data accessible, and the original plan of storing every tweet was rescinded so that only selected tweets are now stored. Critics warn that decades of social history are less accessible in the digital age than prior eras with more durable archival copies of physical books, movies, newspapers and audio recordings. Ironically, digital culture may have less durability than pre-digital culture, making the work of future historians a difficult one. Some types of social media systems vary in their retention policies depending on specific product or user settings. For example, some instant messaging clients do not archive conversations, whereas other clients retain them by default. SnapChat promised users that their video messages will be seen only once (or twice) and will be deleted shortly thereafter. In contrast, some email lists create a searchable archive of prior messages sent to the list (while some others do not by choice). However, it is important to realize that even if there is no centralized archive, individuals at the end points of these services may archive content and make it public at a later date. Such was the case when Usenet content was made easily searchable by Google, upsetting some contributors who never imagined their posts would become easily searchable and available to the masses. People can collect email messages, record Skype calls, log chat sessions, capture screenshots, and collect most digital content

I.  Getting started with analyzing social media networks

18

2.  Social media: New technologies of collaboration

fairly easily. We are now living in a world of easy data collection, retention, analysis, and publication suggesting prudence in using social media systems. Choose your words carefully; they may outlive you.

2.4  Social media examples This section provides a brief description of some popular types of social media as of this writing. Table  2.3 lists the social media systems and categorizes examples of each. It also serves as an index to this section. For a much more comprehensive list of social media tools, see Wikipedia.1 Because of the focus of this book, we highlight the types of networks that these social media tools create by discussing their basic elements and types of connections.

2.4.1  Asynchronous threaded conversation Asynchronous threaded conversations take on many forms such as email, email lists, Usenet newsgroups, discussion forums, and web boards. This form of social media has been the backbone of online communities since before the Internet and it continues to play an essential role in a variety of settings both within organizations and on the public web. Even newer forms of social media such as social networking sites, social sharing sites, and virtual worlds often have asynchronous threaded conversations embedded within them to facilitate discussions (e.g., Facebook groups). Although there are several flavors of threaded conversation, they share some key properties in that they are asynchronous, messages are labeled with an associated address or username, typically, with a single author, and other authors can reply to these messages and reply to others’ replies creating conversation “threads.” Authors do not typically edit one another’s messages or even their own messages after (and some say before) it has been initially contributed. This basic structure has proven to be extremely versatile, supporting a wide range of individual, organizational, and community needs. Threaded conversations, in all their forms, create implicit ties that connect senders and receivers of content in what is often called a “reply network” or “reply graph.” These reply networks can be analyzed to identify important relationships, distinctive patterns of connections that reflect social roles, subgroups or clusters of people, interdepartmental connections, and many other important relationships. Next are brief descriptions of some of 1 https://en.wikipedia.org/wiki/List_of_social_networking_websites

the most important asynchronous threaded conversation systems. See Chapters 9 and 10 for a more complete history and description. Email Email messaging was introduced in the late 1960s and the familiar “@” symbol was introduced by Ray Tomlinson in 1971. Email lists quickly emerged in 1972, the same year as PONG and the year of the last Apollo moon landing. Today, email is almost universal with billions of daily users exchanging hundreds of billions of messages a day. Because of the ubiquity of email, the implicit reply network created by its exchange is often an authentic representation of real-world social connections. In Chapters  9 and 10 we discuss how to analyze three types of email collections: personal email ­collections (e.g., your own email archive), organizational email collections (e.g., your company’s email traffic), and community collections (e.g., email lists). Email lists, discussion forums, Reddit, Quora, and Q&A sites Email lists turn email into a community experience by allowing people to send a message to a single email list address, which is then forwarded to everyone who has subscribed to the list. These collective email exchanges are widely used in enterprise discussion lists or Internet Listservs covering a nearly unlimited array of topics. Email lists facilitate discussions on a topic of interest, technical support, neighborhood gatherings and advocacy, workgroup interactions, internal communities of practice, and even the exchange of goods (e.g., FreeCycle). They are particularly good for reaching less tech savvy users such as older adults who are familiar with email but not more advanced social media technologies. They differ from discussion forums, Reddit, and Q&A sites like Stack Overflow in that they are a “push” technology that shows up in your inbox rather than requiring you to visit a site to get the latest information. Discussion forums emerged before the World Wide Web. In the late 1970s, dial-up bulletin board systems (BBSs) hosted a wide range of message boards that allowed people to post and download information shared on early desktop personal computers. BBS managers selected who could access their services and what content would be retained, exchanged, and copied from other systems. Later, Usenet Newsgroups were created at the University of North Carolina in 1979. Before the World Wide Web, there were tens of thousands of different conversation, each devoted to a variety of topics and containing chains of messages in reply to one another in structures called threads. Usenet newsgroups fostered the collective construction of billions of messages into

I.  Getting started with analyzing social media networks



2.4  Social media examples

19

TABLE 2.3  Types of social media listed with example services Social media type

Examples

ASYNCHRONOUS THREADED CONVERSATION Email

Gmail, Hotmail, MS Outlook

Email lists, Discussion forums, Q&A sites

Listserv, Facebook Groups, Reddit, Quora, StackOverflow

SYNCHRONOUS CONVERSATIONS Chat, instant messaging, texting

IRC, Facebook Messenger, Skype, WeChat, WhatsApp, Slack, GroupMe

Audio and video conferencing

Skype, Zoom, Google Hangouts, Adobe Connect

WORLD WIDE WEB Websites and documents

Faculty member websites, artist portfolio website, Ford.com, umd.edu, SMRFoundation.org, Prevent.org

COLLABORATIVE AUTHORING Wikis

Wikipedia, WikiHow, Docuwiki, Wikia

Shared documents

Google Docs, Zoho, Office 365

BLOGS AND PODCASTS Blogs

WordPress, Tumblr, Medium

Microblogs and activity streams

Twitter, Sina Weibo, Facebook feed

Multimedia blogs, podcasts, and Livestreams

YouTube vlogs, Instagram photo vlogs, iTunes, SoundCloud, Facebook Live, Instagram Live, Twitch

SOCIAL SHARING Video and TV

YouTube, Hulu, Netflix, Vimeo

Photo, images, and art

Flickr, Instagram, Pinterest, DeviantArt

Music

Spotify, Pandora, iTunes

Bookmarks, news, and books

Mix, Reddit, Twitter, Facebook, Goodreads

SOCIAL NETWORKING SERVICES Social and dating

Facebook, eHarmony, Match

Professional

LinkedIn, Zerply

Niche networks

AllTrails, Strava, Untappd, Life Cake, Ravelry

ONLINE MARKETS AND PRODUCTION Financial transaction

eBay, Amazon, craigslist, Kiva, Kickstarter, Indiegogo

User-generated products and services

GitHub, Mechanical Turk, Etsy, fiverr

Review sites

Amazon, Yelp, Angie's List, Google Local Guide Reviews

IDEA GENERATION Idea generation, selection, and challenge sites

IdeaConnection, Chaordix, IdeaScale, Imaginatik, Kaggle, TopCoder

VIRTUAL WORLDS Virtual reality worlds

Second Life, Webkinz, Habbo, IMVU

Massively multiplayer games

World of Warcraft, Lord of the Rings Online, Fortnite, The Sims

MOBILE SERVICES Location and Augmented Reality apps and games

Facebook Checkins, Swarm, Google Lens, Pokemon Go

I.  Getting started with analyzing social media networks

20

2.  Social media: New technologies of collaboration

millions of conversations sorted into tens of thousands of newsgroups. Usenet newsgroups are distinguished from email lists largely in terms of their comparative lack of centralized control and weak boundaries. Anyone could post a message to any newsgroup without regard to membership or the desires of others receiving the messages. Like email and email lists, newsgroups contain a core social network structure called a “reply graph” (see Chapter  10) created when authors are connected with those they reply to in a thread. These early forms of threaded conversations helped inspire innovations that continue to live on in other forms. For example, the Microsoft Research Netscan project demonstrated the value of visualizing social interaction on Usenet, a promise only now being realized to its full potential [9]. Usenet newsgroups also helped inspire one of the first collaborative filtering systems called GroupLens, which made personalized recommendations of content you were likely to enjoy based on the preferences of likeminded people [10]. Since the creation of the World Wide Web, threaded conversations can be found in discussion forums, blog or news comments, Facebook groups, Reddit or Quora discussions, and Question and Answer (Q&A) sites like Stack Overflow. Advanced features provide users with the ability to gain reputations based on the quality of their posts, vote responses up and down (e.g., Stack Overflow responses), review and approve content, report readership levels, recommend related content, and dynamically filter content based on personal preferences to help overcome information overload. The potential cost is the creation of content that is too filtered to admit alternative views. These “filter bubbles” reflect the polarization of political beliefs in many populations.

2.4.2  Synchronous conversation Synchronous conversations such as text messaging, chat, instant messaging, and audio and video conferencing differ from asynchronous conversations in that they occur in real time. Precursors to these Internet-based conversations occurred via telegraph, phone, two-way radios, and similar technologies. UNIX talk messaging, first used in the early 1970s, was among the first instantiations of text-based synchronous conversations based on computer networks. This simple system is the early precursor to chat and instant messaging, initially allowing two people to share a text stream, both users typing characters that would appear intermingled in the same space. Later innovations and refinements separated the text streams and evolved into the many forms of short messaging and texting services available today. Facebook Messenger, Zoom, Apple Facetime, Skype, WeChat, WhatsApp, and other synchronous conversation tools today have added support for small group

c­onversations, whereas others such as Internet Relay Chat enable large scale conversations where anyone who “tunes” to a specific “channel” is able to join. Chat, instant messaging, and texting Chat was one of the most popular forms of interaction on the early Internet and accounted for up to a third of the revenue of the original commercial online providers such as America Online and CompuServe. Inspired by “Citizen’s band” (CB) radio from the 1970s, Chat servers organized discussions into a series of a few dozen “channels” dedicated to a vast array of subjects and interests. The IRC network of “Internet Relay Chat” servers remains a thriving and teeming space filled with chat from many people on numerous topics streaming nearly continuously around the world. More recent forms of chat include platforms like Discord for gamers and Slack for organizations and special interest groups. In commercial chat services, chat channels are frequently policed by the provider’s staff or by appointed volunteers. In the largest noncommercial system, IRC, each channel has an owner who can eject people from the channel, control who enters the channel, and decide how many people can enter. Because of a lack of explicit links tying specific comments to one another, it can be hard for humans and computers to know who is talking to whom. This means that the reply networks created by chat may be error prone or probabilistic in nature. However, networks that connect people together based on who has chatted in the same channels are possible. Instant message text chat clients offer a private, often one-on-one, potentially small group, chat environment. Sometimes referred to as “buddy lists,” these tools allow people to keep a list of their friends and contacts who also use the same or compatible tool. The messenger software indicates which if any of a person’s “buddies” are available and active at that moment for possible conversation (often referred to as “presence”). Selecting a person on your buddy list opens up a private window for exchanging short lines of text, emojis, images, and videos in real time. Tools like Skype, Facebook Messenger, Google hangouts, WeChat and WhatsApp merge the buddy list and text chat with full voice telephony, blurring the distinctions between these modes of communication further. Tools like Slack, Yammer and Chatter bring these types of chat tools into work groups and organizations with a more enterprise focus. Two primary networks connect users of these kinds of services. One network is a friendship network that connects users to the other users on their buddy list. Another network is a conversation network that connects people based on how often they talk with one another. Organizations that provide instant messaging services can use these networks to capture latent and active internal connections.

I.  Getting started with analyzing social media networks



2.4  Social media examples

Text chat has a mobile form called Mobile Short Messaging Service (SMS), which has taken the world by storm, becoming the most widely used form of electronic communication. Trillions of texts are exchanged each year among billions of mobile phone users. Text messaging previously lacked many of the features of richer forms of message exchange, though more are regularly added. It is particularly prominent in some developing countries where mobile phones make up the core communication platform. SMS generates communication networks connecting phones (and their users) to each other based on the number of messages exchanged. For many users of recent mobile devices SMS is often displaced by propriety alternatives like the Apple iMessage and Samsung Messages applications which provide an alternative to SMS with richer media features and stronger security. Audio and video conferencing Audio and video conferencing are highly synchronous forms of social interaction that are often even more interactive and “real time” than text chat alternatives. People interact in near real time, speaking and replying in a nearly continuous loop of verbal exchange similar to face-to-face encounters. Audio conferencing using standard phones has grown steadily to become a widely used service for small teams, as well as for training or marketing sessions for hundreds of users. The simplicity of use, low cost, and emphasis on human voice has turned phone conferences into widely used and productive applications. In an audio conference, neither speakers nor listeners need to worry about their dress, facial expressions, or eye contact, and therefore can engage in other tasks simultaneously without offending others. Audio conferencing now increasingly takes place through Voice over Internet Protocol (VoIP) connections with dramatic reductions in cost and expansion of use around the planet. Tools like Skype and Zoom now make an audio conference among a dozen or more people in as many time zones simple. Videophones have been the promised “vision of the future” since the 1940s but failed to reach mass market adoption for decades. High costs of early system hardware and connections were multiplied by the social awkwardness that video connections impose. Unlike audio only connections, video requires that people comb their hair, straighten their tie, or put on a dress and organize at least the area the camera can see. Video requires a continuous display of engagement, more like a face-to-face encounter, without the flexibility of a phone call that allows wandering attention and multitasking. Despite these hurdles, as hardware and connection costs have plummeted to easy affordability for many and the vision of widely used videoconferencing is now being realized. Inclusion of built-in video cameras and the

21

iChat program on Apple computers and video services in Skype, Facetime, SnapChat, Facebook Live and other instant messaging clients have triggered a much wider community of users, including grandparents seeing distant grandchildren and distributed musical bands and project work teams. Corporate use of videoconferencing platforms like Zoom, Webex, and Adobe Connect has increased steadily and has been integrated with screen and document sharing services that deliver high-quality sound and images. Video cameras on video game consoles and smartphones have further expanded the scope of videoconference users. Social networks are created whenever people connect with one another via audio or video conferencing. However, more fine-grained analysis of the conversations themselves (e.g., who replies to whom) is challenging because the data are difficult to capture automatically. Advances in automated speech to text conversion and video facial recognition may soon make it possible to efficiently automate the extraction of these network exchanges from recordings of video conversations.

2.4.3  The World Wide Web The largest public, machine-readable network is the World Wide Web where web content, such as web pages, and documents, such as images (identified by their Uniform Resource Locator or URL), are connected together by hyperlinks. The World Wide Web, WWW, or just “the web,” was initially conceived by Tim Berners-Lee in the 1980s, but it was not realized until the 1990s. The WWW was the first platform to integrate the concepts of hypertext, developed in the 1960s by Ted Nelson (Xanadu) and Douglas Engelbart (oN-Line System [NLS]), with the Internet [11]. The result was a highly flexible platform that allowed people to view web content hosted on servers throughout the world with the use of a web browser. Today the web is the primary platform upon which most of the social media tools are built. In this section we focus primarily on traditional websites such as corporate, organizational, and government websites, homepages, and documents (e.g., images and pdf files). Although you may not realize it, network analysis already plays a role in billions of people’s everyday lives when they search the web via Google. As the amount of content on the web increased, search engines became essential for making content on the web discoverable. Early search engines looked only within the text of each web page (and its associated metadata fields) to determine its relevance. The first generation of search engines, like Alta Vista, built an index of all the words on millions of webpages and matched them to search queries. Google made a breakthrough in the quality of its results by developing its PageRank algorithm, which

I.  Getting started with analyzing social media networks

22

2.  Social media: New technologies of collaboration

determines how important a page on the web is based on its position in the web’s wider network of connected pages. At its core, the PageRank algorithm views a link to a page as a “vote” for that page’s importance, so that pages with many incoming links score well. In addition, it considers the importance of each of the incoming links: receiving a link from a highly linked-to site counts more than receiving a link from an unknown site. This and related concepts are captured in various measures of the “centrality” of a vertex within a network as discussed in Chapter 6.

of wikis. Collaborative document creation continues to grow in popularity as corporations, governments, and community organizations discover that they can conveniently share and edit documents through Google Docs, Windows 365, or DropBox. Users who view or edit the same documents form a network that may reveal patterns of cooperation, shared interest, or opportunities for new collaborations.

2.4.5  Blogs and podcasts

Wikis The most widely known example of collective document creation is Wikipedia, which is only one example of thousands of smaller wikis scattered throughout the web. Created by Ward Cunningham in the early 1990s, wikis are tools that allow a group of people, potentially any Internet user, to quickly access and edit a shared collection of documents in the form of web pages. There are many “wiki engines” (i.e., wiki platforms) including the open source MediaWiki engine that is used by Wikipedia. Wikis are used to create encyclopedias (e.g., Wikipedia), fan or game sites, corporate intranet content, and information repositories on topics ranging from educational resources to technical documentation to patient support information. Despite many differences among implementations, all wiki engines track each edit of each wiki page, creating a detailed page history. These edits can be reversed, creating a social version of the “undo” function of a word processor. People can follow updates to content on the RecentChanges page, get notified of changes to specific pages after subscribing to (i.e., “Watching”) them, or view people’s “user contribution pages” that chronicle all of their edits. Wikis include many implicit networks. Some networks connect pages to other pages through hyperlinks or connect pages that are grouped together into the same category. Other networks connect people to people when, for example, a person posts on another person’s user page or when two people co-edit the same page. These networks can be mined to better understand content relationships and social roles as described in Chapter 14.

Blogging is a special form of web page publication. Deceptively simple, a blog is essentially a low-cost mechanism for publication of rich digital content. Early blogs presented a series of text messages or “posts” in reverse chronological order so the most recent post was always at the top of the page. Today, a blog is a rich platform for content presentation and commentary. Blogs may contain textual content that are now complemented with pictures, video, and audio. Blogs present this content along with search, functionality for readers to comment on each blog post, tags to categorize posts, pointers to related blogs (i.e., blogrolls), and a range of applications and widgets. Popular tools like WordPress, Drupal, and Blogger make blogging essentially free and widely available. They are also commonly embedded within larger sites. Blogs and by extension bloggers are now often able to build audiences that rival pre-digital media and challenge more established information providers, particularly in the news arena where current information is paramount. Blogs and bloggers are seen as potentially powerful makers and breakers of brands, political candidates, and news stories. They also serve as micropublishing platforms for families sharing stories and photos of their children and niche interest groups exploring an obscure topic. A single blog may be authored by an individual or a handpicked set of authors. Others participate via comments or by linking their own blog to other blogs, creating aggregated collection of interconnected blogs often referred to as the blogosphere. Specialized search engines collect the messages from much of the blogosphere, using the unique properties of blog links (called trackbacks) and number of comments to assign credibility scores to blogs. Because the connections can be automatically captured using web crawlers, researchers have analyzed the blogosphere to better understand issues like the nature of political discourse in the highly divisive politics of the United States [12], as well as other areas such as Iran [13].

Shared documents The idea of collaboration through shared documents such as word processing and spreadsheet files is well established and differs from the community approach

Microblogs and activity streams The Twitter microblogging system gained widespread use in 2007 and has now become a worldwide phenomenon. It is similar to traditional blogs in its focus on recent

2.4.4  Collaborative authoring Several social media tools facilitate the collaborative authoring of documents and repositories, enabling small groups and even communities of thousands to effectively create, maintain, and organize documents and document repositories.

I.  Getting started with analyzing social media networks



2.4  Social media examples

posts, but differs in that its posts, called “tweets,” are restricted to 280 (formerly 140) characters of text. Twitter takes advantage of the idea of blog feeds by allowing you to subscribe to, or “follow,” any other Twitter user. Each user’s feed is personalized to show the most relevant tweets of all individuals he or she is following, creating a live stream of bite-sized information nuggets. A number of competing services exist, such as China’s version called Sina Weibo that boasts nearly 500 million users. Other platforms, such as Facebook and LinkedIn use status messages that serve as microblogs that are broadcast to friends. Microblogging sites create several interesting social network structures. The most obvious network is the one created by the “follows” and “is followed by” relationships. Unlike Facebook, these “follow” relationships are directed: you can follow people who don’t follow you and vice versa. Therefore these connections are not always reciprocated; many connections flow in only one direction. This is in contrast to the undirected or mutual ties present in Facebook friendship connections and LinkedIn. Other networks connect users based on the number of times they reply to others’ microblog posts or repost messages they come across (i.e., “retweet” or RT). A detailed analysis of networks found within Twitter is found in Chapter 11. Multimedia blogs, podcasts, and livestreams As bandwidth and multimedia support has increased, a variety of services related to blogs have appeared including video blogs (vlogs) common on YouTube, photo blogs available on Instagram, audio blogs called podcasts, and livestreams of a person gaming on Twitch or broadcasting video footage on Facebook Live and Instagram Live. Just as a blog privileges the creator of the blog to have primary control and visibility, these multimedia forms also privilege the creator. Multimedia blogs may focus on a specific topic or everyday experiences of a specific individual as a “lifelogging” or “lifeblogging” form of autobiographical journaling. Typically, people can comment in text to the initial posts, and occasionally systems allow multimedia replies, for example a video that replies to another video. Some multimedia sites encourage submissions of content from readers but are vetted by those in charge of the site before being posted. Other multimedia blogs are authored by an individual or small group, which are read by small to large sized groups. Mobile photo blogs like Instagram or Flickr make innovative use of mobile devices such as smartphones to upload photos, videos, and text that is often automatically tagged with location information. Podcasts may include audio or video content and like traditional blogs can be subscribed to using tools like iTunes so that new content is automatically updated or downloaded. They differ in that the facilities for

23

commenting on podcasts are not as common, although they may be provided as part of a website. Technological improvements will likely make search tools and annotation of multimedia content such as videos, images, and audio more common in the future. Livestreaming services, such as Twitch and Facebook Live, allow people to display a live screencast (e.g., of gameplay) or video feed, to which observers subscribe. They typically provide tools to let observers interact with the broadcaster in real time and may record the experience for later consumption. The networks created by multimedia blogs, podcasts, and livestreams are similar to those created by blogs and microblogs. They connect content to content and, by extension, content authors to content authors. People also are implicitly connected to one another when they read the same content, comment on the same blog posts, use the same tags, or even post from the same locations.

2.4.6  Social sharing Social sharing sites are designed to allow individuals to share content, typically of a certain type (e.g., videos, photos, websites). They provide an alternative to purposeful searching for content on search engines by allowing a community of peers to collaboratively identify and share interesting content. They are a modern incarnation of browsing, where the masses decide what is on the shelf. While some services are focused almost solely on social sharing (e.g., Mix, Goodreads), social sharing features often show up as a feature on other social media platforms (e.g., photo sharing on Facebook). Social sharing sites may or may not allow users to create content. For example, YouTube allows anyone to upload videos, while Hulu users can only share what they’ve been watching. Video and TV Since the widespread use of digital video cameras, people have been uploading their videos to share with others. Sites like YouTube and Vimeo allow the masses to easily upload and share video content and link to it or embed it within other websites such as blogs. Corporations, universities, and media outlets often post content on their own YouTube “channels.” Amazon Prime, Hulu, and Netflix streaming video services allow users to search for, view and review TV shows, movies, and shorter video clips. These sites are home to a number of social network structures. For example, YouTube allows users to become “friends” or contacts with one another. In addition, relationships can be created between users when they comment on one another’s videos, make a video a “favorite,” or subscribe to a user’s stream of uploaded videos. Networks of videos that relate to one another are also created based on having

I.  Getting started with analyzing social media networks

24

2.  Social media: New technologies of collaboration

shared tags or shared viewers. A detailed analysis of networks found within YouTube is found in Chapter 13. Photo, images, and art Since the invention of cameras, people have shared photos via albums, scrapbooks, and fridge magnets. Likewise, art is largely created so it can be shared. Social media tools enable the sharing of photos and digital copies of artwork with a select group of other people or the world at large. Pinterest, which allows users to create and share collections of images (i.e., pins), was one of the most successful sites of the past decade, that currently boasts over 250 million monthly active users. Flickr, one of the first image sharing sites, hosts a vast collection of digital photographs that are attached to individuals (e.g., photographers who upload them), groups, and tags that describe them. Like most social sharing sites, Flickr allows users to create networks of contacts (like friends in social networking sites) and limit the distribution of photos to just those individuals or to the world at large. Other similar services exist, such as Google Photos, SmugMug and Amazon Photos, and some sites like SlideShare that allow you to upload slideshow presentations on a wide range of topics. There are also a number of stock photo and vector art sites like iStockphoto that allow individuals to purchase content one image at a time. Sites like DeviantArt allow creators to share and comment on their original pieces of art. Additionally, sites like Facebook include photo sharing and tagging elements. The richly annotated content enables the construction of many types of networks. Some networks connect people who appear in photos together, whereas other networks connect people who follow others’ art or are in the same group. Implicit networks connect people who use similar tags, comment on others’ photos, favorite others’ photos, take pictures in similar locations, or repin images from others. Music A number of social sharing sites revolve around music, including sites like Spotify, Last.fm, and Pandora. These sites share many properties with video and photo sharing sites such as the ability to friend others, post comments, and navigate the site via various metadata fields such as tags and artist. Most sites allow users to create explicit playlists, a modern incarnation of the mix tape, recognizing the value that comes from curating just the right collection of songs. Music sites use collaborative filtering technologies to help recommend music that you are likely to enjoy based on the music you currently enjoy. Collaborative filtering tools, a subset of recommender systems, use data from other users’ behavior to create personalized recommendations. If you like songs that a certain group of users also likes, then you will

probably also like songs they like that you have not yet heard. The networks created by music sharing sites are similar to those of other digital object sharing sites. Bookmarks, news, and books As users review content on the World Wide Web, it is common to want to save a web page, news story, or pointer to a physical book for later use. These collections of pointers can be valuable for other people as well. Several services have emerged to allow users to save a bookmark pointer to a website and share that link with others. Early services like del.icio.us and Digg allowed users to share and rank bookmarks or news stories. While they no longer exist, at least in their original form, they helped inspire key features of services like reddit and tumblr, which provide a range of tools for users to collect pointers to useful and interesting material on the web, annotate it in various ways, and publish it to select others or the public. Users of these sites can filter, search, and sort the accumulated links from many other users. Many users are eager to recall useful material on the web and are often willing to signal their interest or appreciation for certain websites to others. Sites such as GoodReads support sharing book recommendations by adding a social network mechanism to the process to further sort quality content, based on what your friends and contacts like. These services have developed a strong following of users who want to signal their interest in books and authors, which are then aggregated and ranked for others to see. Such services provide a rapid way to identify novel and interesting information, build a historical trace, and form communities of shared interest. Similar sites exist for journal articles and academic research (Academia.edu, ResearchGate). These systems include similar networks to other content sharing sites, as well as networks based on citation linkages created when two or more people co-author a publication.

2.4.7  Social networking services In 1971, Les Earnest wrote the “finger” program that allowed users of a system to check on another user’s status. When a user requested it through the finger program, a file named “.plan” would be displayed to other users. This file soon became the business card and office door for many early users of the Internet and the networks that preceded it. Some users even updated the file regularly to note their current location, activity, and state of mind. This simple status and profile system evolved over time and inspired the creation of systems that allow people to present themselves to others. Modern incarnations, called social networking services, allow people to share contact information, text, images, and videos about themselves with their self-identified friends or followers. Early popular examples like Friendster and

I.  Getting started with analyzing social media networks



2.4  Social media examples

MySpace established the basic outlines of the social network feature set in which users “friend” other users, enabling them to share content and receive updates about each others’ activity. Social and dating Facebook has come to dominate social networking services, even though it is just a decade and a half old. With over 2.3 billion monthly active users by the end of 2018, Facebook contains the largest machine readable “social graph” on earth. There are many ways people connect to one another in Facebook, from the obvious “friending” that starts a Facebook relationship, to the many ways people can subsequently interact by writing on one another’s “wall,” indicating that they “like” other people’s content, sending messages, tagging photos, and joining common fan pages or groups. Facebook and related systems are rich sources of social network data as a result. Many of these social networking services, such as Facebook, impose restrictive terms of use for their data. In contrast to the mostly open and free model that surrounds Twitter, Facebook constrains what data can be accessed and the duration that the data may be used. As a result, analysis of Facebook social networks can be challenging. In the wake of scandals about the use of Facebook data about hundreds of millions of people for targeted political advertising, Facebook has made access to data about activity on its platform even more restrictive. Individuals may extract some Facebook data related to their own interactions and social network, but even that data may only be used for short periods and for specific purposes. See Chapter  12 for examples of how to analyze Facebook networks. Professional Services like LinkedIn provide a social network feature set tuned to the self-presentation of career professionals engaged in business networking. Users can post their resume, receive and send targeted job invitations, recommend co-workers, introduce a colleague to another colleague, exchange private messages, and join groups such as university alumni associations or special interest groups. These networks are becoming a vital part of the job search process in many industries. Niche networks A number of niche social networks have emerged to help people with a common interest connect with one another. These social networks have the advantage of customizable tools that allow members to share information specific to the niche topic, as well as the advantage of having a self-selected group of enthusiasts. For example, AllTrails supports hikers and mountain bikers who can create custom trail maps, rate others’ maps, and use them while on the trail. Strava, for runners and

25

bikers, shares information on each run or bike ride with friends who can comment on them. It also includes a leaderboard for different “segments” (i.e., pre-defined segments of popular trails), adding a competitive component. The Untappd community focuses on social drinking with features that help you rate and review drinks, venues, and even ask for a ride home. Life Cake allows families to share private memories, view content in a timeline format, and create photobooks. Social network data is similar to data available from Facebook, but also includes connections to the niche content, such as the network of Strava users who have run the same segments or the network of Untappd members who have visited the same brewery.

2.4.8  Online markets and production Many social media sites facilitate the creation, evaluation, funding, and exchange of goods and services. Financial transactions Networks of exchange have always been at the core of marketplaces where buyers and sellers meet, exchange news, and make trades, purchase goods, or form plans for future activities. There has been enormous demand for online marketplaces in the form of auction sites such as eBay and Amazon or advertising sites like craigslist that facilitate location-specific ads for products, services, apartments, and jobs. These services generate communities of buyers and sellers who share an interest in the same products. Many small businesses and professionals, such as artists, craftspeople, or photographers, routinely advertise their products through personal or collective websites, along with service providers, consultants, and personal trainers. These independent small businesspeople can reach a wide audience and develop credibility through reputation system tools like eBay’s feedback mechanism. These marketplaces create networks that connect sellers and buyers through transactions, creating a trade network. Related services allow the financial support of projects that might not otherwise be funded. For example, Kiva allows users to donate money to entrepreneurs in developing countries, facilitating microloans, and then follow their progress via blog posts and public repayment statistics. Kickstarter and related crowdfunding sites like Indiegogo allow people to financially support new ideas for products or services. If a critical amount of funds are raised by donors, then producers are obligated to provide their promised products. Finally, a growing number of services called “prediction markets” allow people to buy and sell assets whose cash value is tied to a future event (e.g., who will be the next U.S. president). The market prices are interpreted as the probability of the event occurring. Services like the Iowa Electronic Market

I.  Getting started with analyzing social media networks

26

2.  Social media: New technologies of collaboration

provide financial tools for people to bet on uncertain future events, in aggregate generating information about the “wisdom of the crowd.” All forms of online banking and payment transaction services create a wealth of network data based on who pays whom how much when. The resulting purchasing patterns can be used to understand market dynamics, or as the basis for recommender systems like those found at Amazon. User-generated products A host of social media sites focus on collaboratively developing, sharing, or selling products. The open source software movement is an excellent example, where users contribute code to develop software tools that are then made freely available. Sites like Sourceforge and GitHub provide tools to support developer communities by tracking changes to the software, monitoring the number of downloads, and providing basic discussion capabilities. Amazon’s Mechanical Turk provides a platform for supporting a host of “human intelligence tasks” such as classifying items and identifying images. This “marketplace for work” allows people from around the world to perform these tasks for micropayments that can add up over time. Other communities focus on a specific type of product. For example, Etsy focuses on hand-made crafts; Threadless, focuses on purchasing T-shirts from user-created designs; and fiverr allows users to request video production services from amateurs. These sites generate trade networks, as well as networks that connect people who work on similar projects or use similar tags to describe their products. Review sites Many social media sites allow people to post reviews of products or services. Some sites like Amazon support written reviews as well as ratings of almost every conceivable product. Local versions of review sites, such as Yelp, Google Reviews, and Angie’s List, focus on locationbased services such as restaurants, shopping, and nightlife, or service providers such as doctors, contractors, and service professionals. These sites create explicit networks when people friend each other (e.g., on Yelp), as well as implicit networks when users favorite or review the same places or services.

2.4.9  Idea generation Organizations are increasingly looking for ways to benefit from the collective intelligence of the masses. Several social media sites use “idea generation” tools to help solicit and evaluate new ideas. Companies like IdeaConnection allow organizations to post proprietary challenges to a community of problem solvers. If someone solves the problem, that person is awarded a specified dollar amount. More domain-specific ­ examples

include Kaggle and TopCoder where users compete against each other for prizes in data analysis tasks or coding tasks. These sites create networks that connect people based on shared projects and challenges. If nobody solves the problem, no money is exchanged. Other tools by companies like Chaordix and IdeaScale allow users to post ideas and vote on others’ ideas, helping the best ones bubble to the top. These services create networks that connect people based on who voted on whose ideas. They also create networks that connect ideas to other ideas based on the number of people who liked both ideas.

2.4.10  Games and virtual worlds Virtual worlds, graphical worlds, and massively multiplayer games attempt to model physical places as well as face-to-face interaction. Modern virtual worlds allow users to build new spaces, create objects, and use powerful programming languages to automate their behavior. These sophisticated forms of social media create remarkably rich collections of networks. Even services offering relatively simple game experiences like card games and backgammon offer sophisticated ways of creating friend networks, teams, and rankings. Game systems commonly allow users to create affiliation networks when players join clubs, guilds, tribes, or teams. Within the game play are other processes that create networks as records are created when users shoot or kill one another or trade less lethal materials. Virtual reality worlds Although many multiplayer games continue to focus on combat role playing, many “social” virtual worlds have become a means for widely dispersed groups to maintain personal contact. These include systems like Second Life, The Sims, and IMVU designed for adults, and popular systems designed for children and youth such as Webkinz and Habbo. Virtual worlds typically offer a range of traditional communication channels, as well as the ability to manipulate “avatar” bodies that pose near one another. Like text chat, these systems support synchronous communication. Virtual worlds allow a number of people who occupy the same “room” to meet and “talk” by speaking, posing, gesturing, and sending lines of text or shared spatialized audio conferencing with one another. Because this interaction happens in real time, all the participants must be active at the same time. But in return, virtual worlds provide a powerful sense of social and physical presence that is absent in asynchronous media. Virtual worlds often support simulations of the multichannel quality and nuances of face-to-face interaction by integrating lines of text with gesture, pose, and voice. For children and youth, they provide an engaging environment where users can earn

I.  Getting started with analyzing social media networks



27

2.5  Practitioner’s summary

virtual cash by playing games or completing virtual jobs and use their virtual cash to decorate their virtual home or feed their virtual pets. Virtual worlds and the social data created in them are typically owned by the company that provides them. Thus, owners of virtual world servers have had monopoly control over their systems making it hard to access data for analysis purposes. However, when data is available, it is a rich source of network connections, which are created in virtual worlds whenever users exchange text or virtual items, are near one another, or interact with the same objects. Massively multiplayer games Massively multiplayer online games (MMOs) are video games that include hundreds or thousands of players who interact simultaneously in a persistent virtual world. They utilize the Internet, although they may be played on a computer or game console such as an Xbox or PlayStation. There are many types of MMOs including role-playing games (e.g., World of Warcraft, Everquest), strategy games (Mankind, War of Legends), first-person shooter and survival games (Fortnite), racing games, etc. They may take place in fantasy worlds or virtual worlds that correspond with specific locations such as World War II battlefields or cities. Many MMOs include complex social arrangements such as guilds, tribes, or teams; sophisticated collaboration tools including live audio-feeds; and virtual currencies that allow players to purchase items needed to complete quests or build their empires.

2.4.11  Mobile services The next step from virtual reality is into physical reality. Recent waves of social media tools integrate hardware and software tools to enable users to annotate physical locations. Smartphones track their location via Global Positioning System (GPS), cell tower location services, wireless networks, and beacons, supporting experiences tied to users’ location in space. Once the device has an accurate sense of its location, that information can be associated with all the digital objects created with the device. Photos can be easily linked to the place on earth where they were taken. Restaurant reviews can be associated with a map of the location of the restaurant. A comment can be linked to where it was made. In addition, mobile devices lead to the creation of more social media content because it can be created and captured in spare moments and when notable events unfold. Location is key to social media services that want to provide contextual information about the world immediately around users. Increasingly, this is done via Augmented Reality (AR) features that map virtual objects (e.g., images) and information onto real-world

objects and scenes. Pointing a camera at an object can now reveal relevant information. Furthermore, a host of additional sensors provide functionality that can be called upon by new social media services. These include still and video cameras, audio and motion sensors, accelerometers that detect motion, and tools that enable device-to-device connections. This combined functionality and the proliferation of smart devices promises a bright future for mobile social media tools. Location and augmented reality apps and games Many services allow users to share their current location with friends, such as Google Maps, Apple’s FindFriends, and Swarm. Apps like Foursquare and the now popular Facebook checkin feature allow users to report their location at certain points of interest. In some cases, they can leave virtual messages or gifts for others, write reviews, and earn points or status symbols for checking into locations the most. Many mobile apps now allow users to find people “around” them based on proximity. Augmented reality apps, such as Google Lens allow users to point their camera at a physical object, such as a bridge or statue, and find out information about it. Location and AR games have moved from the fringes to mainstream with the huge success of Pokemon Go. Earlier examples, which still persist, include mobile, social games like Geocaching and Letterboxing that encourage people to hide “caches” that often include small awards or stamps that other players can find, often using GPS tools. Games like Ingress and Pokemon Go encourage players to check in at locations, collect items, and battle virtual creatures. They are highly social games, where players congregate to battle together, recognize usernames of players in their proximity, and work together to achieve common goals. These location sharing and annotation services often contain network structures similar to those found in other social media services, but with geographical location as an additional dimension. Place joins the set of other entities found in many social media services like people, tags, dates, and connections. This allows you to create networks that connect people to each other based on who is within a certain distance or who frequents the same locations.

2.5  Practitioner’s summary Social media tools have become ubiquitous, despite their relatively recent development. The way people have appropriated these technologies has transformed business practices, family ties, and politics in fundamental ways. The impact of social media is complicated, leading to both positive and negative effects, suggesting

I.  Getting started with analyzing social media networks

28

2.  Social media: New technologies of collaboration

the need for more systematic methods for analyzing and understanding social media environments. The past decade has seen a the rise of corporate social media bohemouths including Facebook, YouTube, Reddit, and Twitter, which have added additional features that often appeared originally in novel social networking tools. However, a thriving ecosystem of niche social networking services and tools continues to appropriate novel technologies such as augmented reality, location services, new funding models, and micro-contributions. There is no agreed-upon taxonomy of social media tools or characteristics. Yet it is essential that practitioners spend time understanding which services and features match their personal and professional needs. Corporate and government decision makers who are seeking to use social media for advertising and promoting their products and services will be more successful if they learn which mechanism best reaches their desired audience, and what constitutes acceptable etiquette in those communities. As a starting point, we discussed the following six key dimensions that help characterize social media systems:

govern social media communities [19], organize activities to meet specific goals, find the limits of scalability, and develop tools to better visualize and understand social activity. Related issues of trust, empathy, responsibility, and privacy have strong research foundations, which can be helpful to a wide range of practitioners. Addressing these issues will help designers and community managers make well-informed decisions rather than simply relying on intuition and anecdotes. The rapid pace of commercial development offers new challenges to the research community to evaluate the impact of design changes, novel policies, and evolving norms. What forms of recognition or reward are appropriate for different domains? How can communities that involve participants with different expectations, skills, and experience be accommodated? How can malicious behavior be reduced? Can envisioned benefits to health, education, energy, or international development become a reality?

• • • • • •

[1] V. Bush, As we may think, Atlantic Monthly (1945). [2] D. Engelbart, A conceptual framework for augmentation of man’s intellect, in: P.W. Howerton, D.C. Weeks (Eds.), Vistas in Information Handling, vol. I, Spartan Books, Washington, 1963, pp. 1–29. [3] B.  Shneiderman, C.  Plaisant, M.  Cohen, S.  Jacobs, N.  Elmqvist, Designing the User Interface: Strategies for Effective HumanComputer Interaction, sixth ed., Pearson, New York, NY, 2016. [4] E. Hall, The Silent Language, Doubleday Press, New York, 1990. [5] M.  Smith, S.  Farnham, S.  Drucker, The Social Life of Small Graphical Chat Spaces, in: Proceeding ACM CHI 2000 Conference, the Hague, Netherlands, March 2000, ACM Press, New York, 2000. [6] D. Powazek, Chapter 8: Barriers to Entry: Making Them Work for It, in Design for Community, New Riders, Indianapolis, IN, 2002. [7] E. Ostrom, Governing the Commons: The Evolution of Institutions for Collective Action, Cambridge University Press, New York, 1990. [8] R.  Axelrod, The Evolution of Cooperation, Basic Books, New York, 1984. [9] M. Smith, Tools for navigating large social cyberspaces, Commun. ACM 45 (4) (2002) 51–55. [10] P.  Resnick, N.  Iacovou, M.  Sushak, P.  Bergstrom, J.  Riedl, GroupLens: An Open Architecture for Collaborative Filtering of Netnews, in: ACM Conference on Computer Supported Collaborative Work Conference, 10/1994, Chapel Hill, NC, 1994, pp. 175–186. [11] T.  Berners-Lee, M.  Fischetti, Weaving the Web: The Original Design and Ultimate Destiny of the World Wide Web by its author, Harper Business, New York, NY, 2010. [12] L.A.  Adamic, E.  Adar, Friends and neighbors on the web, Soc. Netw. 25 (3) (2003) 211–230. [13] J.  Kelly, B.  Etling, Mapping Iran’s Online Public: Politics and Culture in the Persian Blogosphere, Berkman Center Research Publication No. 2008–01, Available at: http://cyber.law.harvard. edu/publications/2008/Mapping_Irans_Online_Public, 2008. [14] C.R. Sunstein, C. R, Republic: Divided Democracy in the Age of Social Media, Princeton University Press, Princeton, NJ, 2018. [15] Z.  Tufekci, Twitter and Tear Gas: The Power and Fragility of Networked Protest, Yale University Press, New Haven, CT, 2017. [16] R.E. Kraut, P. Resnick, Building Successful Online Communities: Evidence-Based Social Design, MIT Press, Cambridge, MA, 2012.

Size of producer and consumer population Pace of interaction Genre of basic elements Control of basic elements Types of connections Retention of content

In addition, we briefly introduced some of the more popular social media systems and features, which are outlined in Table  2.3. In doing so, we highlighted the types of networks each of them create, laying the groundwork for the rest of the book, which will discuss how to gain actionable insights from the analysis and visualization of those networks.

2.6  Researcher’s agenda The widespread adoption of social media tools has begun to usher in a golden age of social science research. Social media systems provide a wealth of data about communication patterns, location information, friendships, and other social arrangements. Mining this data is bound to provide numerous insights into human nature for decades to come. There are also many important questions that need answering to help us effectively utilize social media tools to achieve our goals. For example, we need to understand how to support democratic societies in the midst of increasingly divided clans [14], examine the use of social media supported political protests [15], develop ways to build community [16], understand the power dynamics at play in social media, motivate voluntary participation [17], develop persuasive systems [18],

References

I.  Getting started with analyzing social media networks



Additional resources

[17] K.  Ling, G.  Beenen, P.  Ludford, X.  Wang, K.  Chang, X.  Li, et  al., Using social psychology to motivate contributions to online communities, J. Comput. Mediated Commun. 10 (4) (2005) 10. [18] B.J.  Fogg, Persuasive Technology: Using Computers to Change What We Think and Do, Morgan Kaufmann, San Francisco, CA, 2002. [19] J.  Preece, Online Communities: Designing Usability, Supporting Sociability, John Wiley & Sons, Chichester, 2000.

Additional resources D. Boyd, It's Complicated: The Social Lives of Networked Teens, Yale University Press, New Haven, CT, 2014. Democracy Stronger, and Citizens more Powerful, Brookings Institution Press, Washington, 2009.

29

D.  Easley, J.  Kleinberg, Networks, Crowds, and Markets: Reasoning About a Highly Connected World, Cambridge University Press, New York, 2010. C. Fuchs, Social Media: A Critical Introduction, Sage, Los Angeles, CA, 2017. L. Rainie, B. Wellman, Networked: The New Social Operating System, MIT Press, Cambridge, MA, 2012. C. Shirky, Here Comes Everybody: The Power of Organizing Without Organizations, The Penguin Press, New York, 2008. M. Smith, P. Kollock (Eds.), Communities in Cyberspace, Routeledge, London, 1999. S. Turkle, Alone Together: Why We Expect More From Technology and Less From Each Other, Hachette, UK, 2017. E.  Wenger, N.  White, J.D.  Smith, Digital Habitats: Stewarding Technology for Communities, CPsquare, Portland, 2009.

I.  Getting started with analyzing social media networks

C H A P T E R

3 Social network analysis: Measuring, mapping, and modeling collections of connections O U T L I N E 3.1 Introduction

31

3.2 The network perspective 3.2.1 A simple twitter network example 3.2.2 Vertices 3.2.3 Edges 3.2.4 Network data representations

32 33 34 34 34

3.3 Types of networks 3.3.1 Egocentric, partial, and full networks 3.3.2 Unimodal, multimodal, and affiliation networks 3.3.3 Multiplex networks

36 36

3.4 The network analysis research and practitioner landscape 3.5 Network analysis metrics 3.5.1 Aggregate network metrics 3.5.2 Vertex-specific network metrics 3.5.3 Grouping, clustering, and community detection algorithms

36 37 37 39 40 40

42

3.6 Social networks in the era of abundant computation

44

3.7 The era of abundant social networks: From the desktop to your hand

47

3.8 Tools for network analysis

47

3.9 Node-link diagrams: Visually mapping social networks

48

3.10 Common network analysis questions applied to social media

48

3.11 Practitioner’s summary

49

3.12 Researcher’s agenda

50

References

50

Additional resources

51

41

3.1 Introduction Human beings have been part of social networks since our earliest days. We are born and live in a world of connections. People connect with others through social networks formed by kinship, language, trade, exchange, conflict, citation, and collaboration. Using computer technologies to create social networks is relatively new, but networks of social interactions and exchanges are primordial. Simply defined, a network is a collection of things and their relationships to one another. The “things” that are connected are called nodes, vertices, entities, and in some contexts people. The connections

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00003-0

3.5.4 Structures, network motifs, and social roles

between the vertices are called edges, ties, relationships or links. Many natural and artificial systems form networks, which exist in systems from the atomic level to the planetary level. A special subset of networks are social networks which are created whenever people interact, directly or indirectly, with other people, institutions, and artifacts. Social network theory and analysis is a relatively recent set of ideas and methods largely developed over the past century. It builds on and uses concepts from the mathematics of graph theory, which has a longer history, starting with Leonhard Euler in 1736. Using network analysis, you can visualize complex sets of relationships as maps (i.e., graphs or sociograms) of

31

© 2020 Elsevier Inc. All rights reserved.

32

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

connected symbols and calculate precise measures of the size, shape, and density of the network as a whole and the positions of each element and group of elements within it. The recent proliferation of Internet social media applications and smartphone devices has made social connections more visible than ever before (Chapter  2). A new subset of social networks, social media networks, are a growing focal point for the application of network analysis tools. The idea of networks, whether they are composed of friends, ideas, or web pages, is increasingly an important way to think about the modern world. You can use social network analysis to explore and visualize patterns found within collections of linked entities that include people. From a social network analysis perspective, the treelike “org-chart” that commonly represents the hierarchical structure of an organization or enterprise is too simple and lacks important information about the cross connections that exist between and across departments and divisions. In contrast with the simplified tree structure of an org-chart, a social network view of an organization or population leads to the creation of visualizations that resemble maps of highway systems, airline routes, or rail networks (see Chapter 9). Social network maps can similarly guide journeys through social landscapes and tell a story about how some points or people are at the center or periphery of the network. Maps of transportation networks where distance is measured in number of flights or road miles from one city to another city are familiar. They inspire application to less familiar networks of electrical connections, protein expression, and webs of information, conversation, and human connection. Social network analysis and metrics are described in several excellent books and journals [1–6]. This chapter touches on the key historical developments, ideas, and concepts in social network analysis and applies them to social media network examples. We have left details of advanced topics and mathematical definitions of various concepts to the many fine technical works. The following is intended as an introductory survey of the core network concepts and methods used in subsequent chapters, which focus on the networks that can be extracted from social media sources like Twitter, Facebook, email, discussion forums, YouTube, and wikis.

3.2  The network perspective Network analysts see the world as a collection of interconnected pieces. Those studying social networks see relationships as the building blocks of the social world, each set of relationships combining to create emergent patterns of connections among people, groups, and things. The focus of social network analysis is between, not within, people. Whereas traditional

social science research methods such as surveys focus on individuals and their attributes (e.g., gender, age, income), network scientists focus on individuals and their “alters”—the people to whom they connect. Network analysis shifts the focus of analysis to the bonds between individuals in addition to the internal qualities and abilities of individuals. This change in focus from attribute data to relational data dramatically affects how data are collected, represented, and analyzed. Social network analysis complements methods that focus more narrowly on individuals, adding a critical dimension that captures the connective tissue of societies and other complex interconnections. Network analysis shares some core ideas with the real estate profession. In contrast to approaches that look at internal attributes of each individual, network analysis shares the real estate focus on location, location, location! The interior of a house may be a liability, but where a property is located matters far more when trying to get a good sale price. The network perspective looks at a collection of ties among a population and creates measurements that describe the location of each person or entity within the structure of all relationships in the network. The position or location of a person or “node” or “vertex” in relation to all the others is a primary concern of social network analysis. Many network explanations look for causes of outcomes in the patterns of connections around an individual instead of their personal characteristics. “Know who” is often more important in network explanations than “know how.” Network approaches observe that different people in similar social positions often act in similar ways, even if they have different backgrounds. Positions within networks may be as significant a factor as any aspect of the people who occupy them. Network analysis argues that explanations about the success or failures of organizations are often to be found in the structure of relationships that limit or provide opportunities for interaction [7]. Many network concepts are intuitive and echo familiar phrases like “friend of a friend,” “word of mouth,” and “six degrees of separation.” Other network terms like “transitivity,” “triadic closure,” and “centrality” (see Section 3.5) may be unfamiliar terms for familiar social arrangements. Many of us recognize social network differences among people: we know some people who are “popular” and have connections to many others. We may also know some people who may be less “popular” but are still “influential,” connecting to a smaller number of people who have “better” connections. Network analysis recognizes these and other less intuitively sensed patterns in social relationships, like measuring the number of your friends who know each other and how much a person occupies a gatekeeper or bridge role between two groups. The network analysis approach makes the web of interconnections that bind people to one another visi-

I.  Getting started with analyzing social media networks



3.2  The network perspective

ble, creating a mathematical and graphical language that can highlight important people, events, and subgroups.

3.2.1  A simple Twitter network example To better understand the network perspective, consider the social network of Twitter users shown in Figure  3.1 (see Chapter 11 for a description of Twitter). It is an example of a sociogram, also called a network graph, which is a common way of visualizing networks. Like all networks, it consists of two primary building blocks: vertices (also called nodes or agents) and edges (also called ties or connections). The vertices are represented by images of the Twitter user profile photo, and the edges are represented by the lines that point from one vertex to another. This network graph visualization paints a picture of the social relationships among the Twitter accounts of members of the United States Senate in 2018. The size of each Twitter user’s profile image is determined by the user’s total number of followers as reported by the Twitter Application Programmer Interface (API), which

33

gives software access to extended details about each user’s profile and message data. This is one example of how attribute data (e.g., data that describe a person) can be overlaid onto a network. A line, or edge, exists between two people when one user account “mentions” or “replies-to” another. All of these connections in aggregate reveal the emergent structure of two large distinct groups (G1 and G2) with relatively few connecting links, which loosely map to the two political parties in the United States. These separate clusters reflect the higher rate that members of one party mention one another in contrast to the rate they mention members of the opposing party. This network analysis identifies the individuals who fill important positions within the network, such as those with whom many other people interact and those who are connected across cluster boundaries. The current and following chapters will provide a guide to creating maps like these from Twitter and other social media platforms and data sources. For now, let’s consider the major components of a network in a bit more detail.

FIGURE 3.1  A NodeXL social media network diagram of relationships among Twitter handles for members of the 2018 United States Senate. The size of each user's vertex is proportional to the number of Twitter followers that user had at the time. https://nodexlgraphgallery.org/Pages/ Graph.aspx?graphID=176627 I.  Getting started with analyzing social media networks

34

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

3.2.2 Vertices Vertices, also called nodes, agents, entities, or items, can represent many kinds of things. Often they represent people or social structures such as workgroups, teams, organizations, institutions, states, or even countries. At other times they represent content such as web pages, keywords, or videos. They can even represent physical or virtual locations or events. Vertices often correspond with the primary building blocks of social media platforms as described in Chapter  2: pages in wikis, friends in social networking sites, and posts or authors in blogs. Although it is not an absolute requirement for network analysis, having attribute data that describe each of the vertices can add insights to an analysis and visualization. For example, Figure 3.1 used descriptive attribute data about the total number of followers for each user to convey a sense of who is most popular on Twitter within the network. Other attribute data from Twitter, such as the number of people each user follows and the date they joined Twitter, can also be mapped to visual attributes (see Chapter 11). More generally, attribute data may describe demographic characteristics of a person (age, gender, race), data that describe the person’s use of a system (number of logins, messages posted, edits made) or other characteristics such as income, location or brand preferences. In network visualization tools like NodeXL, attribute data can be mapped to visual properties such as the size, color, or opacity of each vertex (see Chapter 5).

3.2.3 Edges Edges, also known as links, ties, connections, and relationships, are the connective tissue of networks. An edge connects two vertices together. Edges can represent many different types of relationships like proximity, collaboration, kinship, friendship, trade partnership, literature citation, investment, hyperlink, transaction, or any shared attribute (e.g., people who attended the same University). An edge can be said to exist if it has some official status, is recognized by the participants, or is observed by exchange or interaction between them. In summary, an edge is any form of relationship or connection between two entities. Network scientists have developed a language to describe different types of edges. In Section  2.3.5 of Chapter 2, we introduced the core types of connections that occur in social media networks. Here we describe how those concepts map to network and graph theory concepts more generally. Undirected or directed edges are the two major types of connections. Directed edges (also known as asymmetric edges) have a clear origin and destination:

money is lent from one person to another, a Twitter user follows another user, an email is sent from an author to a recipient, or a web page links to another web page. They are represented on a graph as a line with an arrow pointing from the source vertex to the recipient vertex (see Figure  3.1). Directed edges may be reciprocated or not. If I sent you a message, you may send one back in return, or not. An undirected edge (also known as a symmetric or mutual edge) simply exists between two people or things: a couple is married, two Facebook users are friends, or two people are members of the same organization. No origin or destination is clear in these mutual relationships. They cannot exist unless they are reciprocated. Undirected edges are represented on a graph as a line connecting two vertices with no arrows. Edges can be further described by additional types of data. The simplest type of edge, an unweighted edge or binary edge, only indicates if an edge exists or not. For example, a friendship tie between Facebook users either exists or it does not. In contrast, a weighted edge includes values associated with each edge that indicate the strength or frequency of a tie. For example, a weighted edge between two Facebook users may indicate the number of photo comments exchanged or the duration since the creation of a friendship. Weighted edges are often represented visually as thicker or darker or as more or less opaque lines. Including weighted edge data in a network dataset is preferable because this provides additional information about each tie. However, many social network analysis metrics (see Section  3.5) are designed for unweighted networks. Fortunately, any weighted network can be converted to an unweighted one by choosing a cutoff point. For example, an unweighted edge could be shown between individuals who exchanged at least 10 email messages, with no edge between people who exchanged fewer than 10 messages.

3.2.4  Network data representations Because network data differ from attribute data, a different way to represent it is used. With attribute data, it is common to create a data matrix where each row represents an individual and each column represents an individual’s characteristics, behaviors, or answers to survey questions. A modified approach is used to represent relational data. Like attribute matrices, each row represents an individual in the network. However, unlike attribute matrices, each column represents other individuals as shown in Table 3.1. Different types of edges can be represented in network matrices. Table 3.1 describes a directed network because not all connections are reciprocated. For example, Ann “points to” Bob as shown in row 1, but Bob does not “point to” Ann as shown in row 2. If it were an u ­ ndirected

I.  Getting started with analyzing social media networks



35

3.2  The network perspective

network it would be a symmetric matrix; if Ann points to Bob then Bob must necessarily point to Ann. This network is a binary network because it only includes 1s and 0s, where a 1 indicates that there is a connection and a 0 indicates that there is no connection. Allowing additional values would create a weighted network. For example, the 1s could be replaced with the number of email messages sent or phone calls made to the other person. Notice that the diagonal of the matrix connects each person with himself or herself. In this network, like most networks, the diagonal values are 0 indicating that a person does not “point to” herself. However, in some networks a “self-loop” connecting a person to herself can exist. For example, a person may send herself an email message as a reminder. Network matrices are powerful forms of representation that lend themselves to efficient mathematical manipulation for those inclined. However, they can also become quite large and challenging to navigate, particularly when networks are relatively “sparse” with few connections and many items. TABLE 3.1  A network represented as a matrixa Ann

Bob

Ann

0

1

1

Bob

0

0

0

Carol

1

0

0

Carol

a

This network is a directed network, as it is not symmetrical (i.e., Ann points to Bob in row 1, but Bob doesn't point to Ann in row 2). It is a simple binary network: either a tie exists (value = 1) or not (value = 0).

ADVANCED TOPIC

The foundations of graph theory

An alternative to the matrix data format that is a more efficient representation of a network is called an “edge list.” As its name suggests, it is simply a list of all edges in the network as shown in Table 3.2. This is the same network as shown in Table 3.1. Individuals in the Vertex1 column “point to” those in the Vertex2 column. Unless data describing the value of each edge are provided in additional columns, the network is implied to be a binary one. Self-loops are possible to represent in edge lists by having a row with the person’s name repeated in both columns. Throughout this book, you will use edge lists instead of matrices. Edge lists are “efficient” in that they only record a row of data for each connection that does exist in a network, rather than store a “zero” for each possible connection that does not exist. Edge lists can be smaller files and easier to edit and review. The final method for representing networks is through network graphs. Figure 3.2 is a network graph based on the data in Table  3.2. It makes immediately clear that the relationship between Ann and Carol is reciprocated (i.e., there are arrows on both sides of the line connecting them) and that there is no connection between Bob and Carol. Our earlier analysis of Figure  3.1, another network graph, demonstrates how network graphs can lead to insights that are hard to identify in tabular data, particularly when large networks are presented. However, many network graphs require ­ significant preparation to assure that they are readable as described in Section 3.9 and Chapter 4. TABLE 3.2  A network represented as an edge lista Vertex1

Vertex2

Network analysis is rooted in the work of the mathematician Leonhard Euler who in 1736 studied the question whether a single path could be walked over the Seven Bridges of Königsberg that connected islands in the river Pregel (which flows through what was then Prussia and is now Kaliningrad in Russia) without crossing any bridge more than once.1 By reimagining the problem in terms of vertices and edges, he showed it is impossible to cross each bridge just once. Although the problem seems abstract, its solution led to the development of the mathematics of graph theory and, notably, hundreds of years later, the mathematical work of Paul Erdös and Alfréd Rényi on random graphs in the 1950s, an important theoretical development that allows for the generation of a graph from random processes. Social network analysis builds on these concepts and extends them to capture the nonrandom connections that occur among groups of people.

Ann

Bob

Ann

Carol

Carol

Ann

1 https://en.wikipedia.org/wiki/ Seven_Bridges_of_K%C3%B6nigsberg.

FIGURE 3.2  The directed, binary network described in Tables 3.1

a

Individuals in the Vertex1 column “point to” those in the Vertex2 column in this directed network. The network is implied to be a binary network. Additional columns could be used to describe each edge. For example, an Edge Weight column could be added with values representing the strength of various ties.

Ann

Carol

Bob

and 3.2 represented as a network graph. Arrows indicate the direction of the connection (e.g., from Ann to Bob).

I.  Getting started with analyzing social media networks

36

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

3.3  Types of networks Social networks range in size from a handful of people to national and planetary populations. They also differ in the types of vertices they include, the nature of the edges that connect them, and the ways in which they are formed. In this section we introduce some of the distinctions that network scientists have identified to describe different types of networks. These distinctions affect the metrics and maps generated from them, as well as their interpretation.

3.3.1  Egocentric, partial, and full networks It is often useful to consider social networks from an individual member’s point of view. Network analysts call the individual that is the focus of attention “ego” and the people he or she is connected to “alters.” Some networks, called egocentric networks, only include individuals who are connected to a specified ego. For example, a network of your personal Facebook friends would be an egocentric network because you are, by definition, connected to all other vertices, like the hub of a wagon wheel with many spokes. Other egocentric networks and their associated “subgraphs” (see Chapter 7) may extend out from an ego, reaching not only friends, but also friends of friends. More generally, egocentric networks can extend out any number of “degrees” from an ego. The basic “1-degree” ego network consists of the ego and their alters. The “1.5degree” ego network extends the 1-degree network by including connections between all of the alters. For example, a Facebook 1.5 degree ego network would characterize which of your friends know each other (sadly this data is no longer available from the Facebook platform). The “2-degree” ego network extends the 1.5-degree network by including all of the alters’ own alters (i.e., friends of friends), some of whom may not be connected to the ego. These three sizes of ego networks allow you to look at increasingly larger, but still “local” neighborhoods around a particular individual in a social network. Higher-degree networks (e.g., 2.5, 3) are feasible to create but not used as often in practice because they can quickly grow to a large size and become intractable. Consider, for example, that of the 1.59 billion Facebook users in 2016, there were an average of only 3.57 “intermediaries” between any two people in the network!2 Networks that are smaller than the complete human population are often interesting and some can be small enough to be manageable with the resources

2 https://research.fb.com/three-and-a-half-degrees-of-separation/.

available in a desktop or laptop computer. A “full” or “complete” network contains the subset of people or entities who match some interest or attribute and includes information about the set of connections among them all. All the “egos” in a full network are treated equally, none is assumed to be the “ego” of the network, although analysis of these networks will reveal that some people are more strategically located in the network than others. A full network is often created and available when a single system, such as a social media platform, acts as a hub among a group of connected people. For example, the Twitter network includes all users of the service and the connections between them. In practice, it is not always feasible (or particularly insightful) to analyze a ­platform-scale full network in one dataset. Instead, analysts create more selective sub-networks by selecting a sample or slice of the larger complete network. For example, Figure 3.1 showed the slice of the Twitter network that included the connections among the 100 user accounts for the members of the 115th United States Senate. This partial network is based on a known list of users. Other types of networks are topic centric, they start with a search term and the people who will be included in the data are not (necessarily) known prior to the data collection. Other partial networks may be created to include a subgroup of users (e.g., all conference attendees), or include only people and connections that occurred within a specified time frame, or be limited to people who have certain characteristics (e.g., CEOs of Fortune 500 companies, members of a national or state legislature).

3.3.2  Unimodal, multimodal, and affiliation networks Up until this point we have only considered networks that connect the same type of entity. These standard networks are called unimodal networks because they include one type (i.e., mode) of vertex. They connect users to users or they connect documents to documents, but they don’t include both users and documents. However, networks can include different types of vertices creating multimodal networks. Chapter  6 includes an example multimodal network that connects Marvel Movies to Characters in those movies. Rich sets of intersecting networks often form in social media environments composed of connections between people, photos, videos, messages, documents, groups, organizations, locations, and services. In many cases, these multimodal networks have to be transformed into simpler unimodal networks to perform meaningful network analysis, as most network metrics are designed for unimodal networks.

I.  Getting started with analyzing social media networks



3.4  The network analysis research and practitioner landscape

A common type of multimodal network is a bimodal network with exactly two types of vertices. Data for these networks often include individuals and some event, activity, or content with which they are affiliated, creating an affiliation network. For example, an affiliation network may connect users with the wiki pages they have edited. People are affiliated with pages. In this network, no two users would directly connect to each other. Likewise, no two pages would directly connect to each other. Pages only link to people (i.e., editors). Bimodal affiliation networks can be transformed into two separate unimodal networks: a “user edits page” network can be converted into a user-to-user network and an page to page network (see Chapter 6, Advanced topic: Transforming a bimodal affiliation network into two unimodal networks for details). The user-to-user network connects people based on their indirect links to one another through edits to a common page. For example, in a wiki co-edit affiliation network Derek and Marc would be strongly connected because they both edit many of the same wiki pages. In contrast, a Page to Page network connects Pages based on the number of shared editors. For example, a pair of wiki pages would be closely connected if many people edited both of the pages (see Chapter 14). More generally, this approach can be used to relate objects of all types (e.g., books, photos, and audio recordings) based on users’ behaviors (e.g., purchasing or reading habits) and preferences (e.g., ratings). Affiliation networks are the raw material of many recommender systems that recommend items of interest, such as Amazon’s “Customers Who Bought This Item Also Bought” feature. A network data structure can return results to queries like “people who linked to this document also linked to these documents” or “if you link to this document, you may want to link to these people.”

3.3.3  Multiplex networks Although it is common for two people to be connected in many different ways (e.g., by exchanging phone calls, emails, sharing group membership, and being married), most networks only include one type of connection or edge. However, it is possible to consider networks with multiple types of connections, called multiplex networks. For example, the Twitter network shown in Figure 3.1 includes two types of directed edges: “reply to” relationships and “mention” relationships. The network graph visualization could have uniquely represented each type of edge by using color, different edge types (e.g., dotted lines, solid lines), or edge labels (see Chapter 5). In the case of Figure 3.1, the difference between the two types of edge (reply and mention) was not deemed important, so the multiplex network data was condensed into a uniplex network that showed a

37

single directed edge if one or more of the three types of connections were present. This strategy of combining multiple types of edges is a common one that allows for the use of network metrics, which are mostly based on uniplex networks.

3.4  The network analysis research and practitioner landscape You can find network scientists in nearly every academic discipline and an increasing number of practitioner communities. Network concepts and techniques are now widely found throughout a range of disciplines including sociology, anthropology, communications, computer science, education, economics, physics, management, information science, medicine, political science, public health, psychology, biology, history and digital humanities. In the past several decades, social scientists have shown that network structures have a profound influence on health, work, and community. Getting a job, being promoted, catching an illness, adopting an innovation, and many more activities and processes have been explained in the terms of social networks. Network structures are important in the biological sciences where research is focused on connections between metabolic and genetic processes. The shape and function of networks can have great consequences as ideas, genes, innovations, or pathogens diffuse through populations. Researchers now apply network theory and methods to understanding how Supreme Court decisions relate to previous cases, how the United States Senate votes (see Chapter 7), how epidemics spread within cities, and how characters in a movie relate to one another (see Chapter  6). Networks are formed from many physical processes and are echoed in a number of structures created inside information systems such as the collection of linked documents within the World Wide Web or an enterprise’s collections of files and emails. Information scientists use these links to identify high-quality web pages (e.g., Google’s PageRank algorithm), or use the citations from research articles to identify high-impact articles and authors. Network methods are diffusing beyond academic research, becoming an important tool for managing organizations, markets, and movements. Entrepreneurs apply network analysis techniques to understand how to leverage the powerful effects of word-of-mouth marketing as their customers spread news about their new products to one another. Many politicians recognize the potential power of a connected network of supporters who can be turned into contributors, volunteers, and voters. Engineers use network analysis to build more effective power grids, computer networks, and transportation systems. Law enforcement officers and lawyers

I.  Getting started with analyzing social media networks

38

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

analyze email networks to identify and prosecute potential criminals. And the intelligence community seeks to identify national security threats by looking at networks created by communication links, money trails and kinship. Having at least a basic understanding of network thinking and concepts is a core literacy of our time. Like statistics, network analysis has countless applications to a number of fields.

This book primarily focuses on social network analysis, a subfield of network sciences that focuses on networks that connect people or social units (i.e., organizations, teams) to one another (see Advanced topic: Early social network analysis). Further, we are interested in networks that connect human-generated content or artifacts together, such as websites or cell phones, or social media networks.

A D VA N C E D T O P I C Early social network analysis

The social science roots of social network analysis can be found in the early 1800s in the work of the person credited with being the first sociologist, Auguste Comte, and later in the early 1900s in the work of the sociologist Georg Simmel. Both saw patterns of social ties as the main focus of sociology in contrast to the study of individuals and their attributes. Early in the 19th century, Comte defined society as more than simply a group of people. He argued that a population became a society only when people had influence on one another and considered the choices and interests of others as part of their own choices. Simmel echoed these ideas at the turn of the twentieth century, focusing social science on the study of how people come together and form groups and associations. These sociologists imagined society as composed of a web of relationships—more than a mass of individuals; they saw societies as networks of interaction and influence. The idea of connected actions linking people to one another has remained at the core of the social sciences, but efforts to create a systematic language to record social relationships started only in the 20th century. Anthropologists studying the range of kinship systems they documented in fieldwork from around the world created symbol systems that are related to social network analysis. Their maps of who is related to whom were early forms of social networks focused on just the subset of social ties that are considered to be “family.” The core concepts and methods of modern social network analysis date from the 1930s and the pioneering work of Jacob Moreno and his many collaborators. Researchers at New York University, Columbia, and Harvard created the first scholarly works featuring the distinctive core components of modern social network theory: measures, maps, and models. Moreno and his research partners created the first pictures of patterns of groups of people and their partnerships, using visual maps with symbols that represented individuals with different types of lines connecting them to others that represented different kinds of relationships.

Moreno documented relationships among schoolchildren and the way an innovative behavior, running away, moved through chains of student connections. In 1934, Moreno [8] published “Who shall survive,” which catalyzed work among a group of scholars who refined his approach and added critical mathematical elements that today are a standard part of network analysis. These approaches were applied to various settings, and revealed the key roles a relatively small number of people played in their networks along with the presence of subgroups of distinct people. For example, in the 1930s, Davis et al. collected detailed records of observed attendance at 14 social events by 18 southern women, and the graph of that data revealed two distinct groups with minimal overlap [9]. Moreno developed sociometry and is often considered the founder of the sociogram, applying these diagrams in studies of relationships among members of a football team. These diagrams revealed patterns of friendship and animosity (see Figure  3.3) (as produced in Freeman [10]). At Harvard in the 1930s, a group formed around W. Lloyd Warner and Elton Mayo to explore interpersonal relationship in workplaces. Early social network analysis work focused on connections in small work groups in industrial factory settings. For example, Roethlisberger and Dickson [11] studied the Western Electric Wiring room, documenting the ways individuals within a group worked with one another. As seen in Figure 3.4, some workers in the study emerged as the most connected, whereas others appeared as peripheral or isolated. Another dataset was created that represented the relationships among 14 manufacturing employees of the Western Electric Hawthorne Plant. Employees and two inspectors were observed, and each contact among them was coded. When employees played games with one another, argued, were openly friendly, confrontational, or helpful a note and tie was recorded. The result were six networks, which led to a seminal work by the Harvard sociologist George Homans [12] and later more mathematical work that focused on

I.  Getting started with analyzing social media networks



39

3.5  Network analysis metrics

I3

I1

W1

W2

W2

W4

S1

W5

S2

W6

W7

W8

W9

S4

FIGURE 3.4  An early social network diagram of relationships among workers in a factory illustrates the positions different workers occupy within the workgroup. From Management and the Worker: An account of a Research program conducted by the western electric company, Hawthorne works, Chicago by F. J. Roethlisberger and William J. Dickson, Cambridge, Mass.: Harvard University Press, Copyright © 1939 by the President and Fellows of Harvard College. Copyright renewed © 1966 by the President and Fellows of Harvard College.

FIGURE  3.3  Jacob Moreno's early social network diagram of positive and negative relationships among members of a football team. Originally published in Moreno, J.L., 1934. Who Shall Survive? Nervous and Mental Disease Publishing Company, Washington, DC.

­automatically finding clusters or groups within these datasets [13]. In the 1950s, Nadel wrote about social roles and the social structures that define them [14]. He saw that the patterns of connections people had might be similar, even

3.5  Network analysis metrics Social scientists, physicists, computer scientists, and mathematicians have collaborated to create novel theories and algorithms for calculating measurements of social networks and the people and things that populate them. These quantitative network metrics allow analysts to systematically inspect the patterns of connection within the social world, creating a basis on which to compare networks, track changes in a network over time, and determine the relative position of individuals and clusters within a network. Social network measures initially focused on simple counts of connections and over time became more sophisticated as it developed and incorporated concepts of network density, centrality, structural holes, balance, and transitivity. Some metrics describe a network as a whole. For example, vertex count is the number of entities in the network while the edge count is the number of connections among them. Another whole network metric “density”

if they were connected to different people. These patterns, Nadel suggested, could be studied systematically, but in the 1950s the data and computational resources made that ambition a challenge. Over time, Moreno's colleagues, including Paul Lazarsfeld, added key ingredients of the modern form of social network analysis: metrics and algorithms for calculating important network properties of the graph as a whole and for each individual in the graph (see Freeman [10] for details).

captures how connected a set of vertices are by calculating the percentage of connections that are observed from maximum possible count if everyone connected to everyone. Other metrics are calculated for each vertex in a network. For example, “centrality” measures, of which there are many, capture how “important” (central) a vertex is within the network based on some objective criteria. Some people sit at the edge or periphery of their networks, whereas others are firmly at the center, connected to many of the other most connected people. In most human networks, even highly connected networks, some pairs of people are not directly connected. When a third person bridges a connection (a “friend of a friend”), we can think of that person as a broker, a “bridge” or a “connector.” When that person is missing, we can think of the gap as a “structural hole,” a place in which there is a missing connector, potentially a good spot to build a “bridge.” The following sections describe some of these metrics in more detail. Chapter 6 introduces some of the core metrics found in NodeXL through hands-on exercises.

I.  Getting started with analyzing social media networks

40

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

ADVANCED TOPIC Historical obstacles to the development of network analysis Following the rapid development of the major elements of social network analysis in the 1930s there was a period of stagnation and neglect. For a variety of reasons, from Moreno’s own personal and professional conflicts to the cost and lack of available network datasets and computing resources, social network analysis languished for decades. The early social network literature was built on manually collected and processed data about social ties. Researchers would typically observe or survey population members, asking each to list those they came in contact with regularly for a variety of tasks and purposes. People are often unable to recall all their interactions accurately. The prohibitive cost of this approach was also a major limiting factor in the widespread application of social network analysis in enterprises and organizations. The recent explosion of computer-mediated social relationships and the associated drop in the costs of creating network datasets have made network approaches increasingly practical. As more details about our interactions and associations are tracked and captured by mobile devices and social media services, network analysis becomes increasingly useful. Network analysis is computationally intensive: many network metrics can require generating millions of calculations even when processing modest sized datasets. The recent explosion of computing power and the associated drop in costs have made network approaches increasingly practical, even if network methods remain among the most computationally intensive in use.

3.5.1  Aggregate network metrics A number of metrics are used to describe and summarize an entire network. In some cases, a single network dataset contains sub-networks separated into several disconnected pieces, called components. Some aggregate network metrics only work on networks where all of the vertices are connected in a single component, whereas others can be applied to entire networks even if they are split up into disconnected segments. Here we describe just a few aggregate network metrics to give a flavor for what is possible, leaving a fuller discussion for Chapter 6. As mentioned, density is an aggregate network metric used to describe the level of interconnectedness among a set of vertices. Density is a count of the number of relationships observed to be present in a network divided by the total number of possible relationships that could be present. It is a quantitative way to capture important sociological ideas like cohesion, solidarity, and membership.

Centralization is an aggregate metric that characterizes the amount to which the network is centered on one or just a few important nodes. Centralized networks have many edges that emanate from a few important vertices, whereas decentralized networks have many vertices with many interconnections. Networks with high levels of centralization are likely to be more hierarchical, with a few people playing hub roles. Other metrics integrate attribute data with network data. For example, metrics that measure homophily look at the similarity of people who are connected. Studies typically show that people are connected to others who are similar to themselves on core attributes like income, education level, religious affiliation, and age.

3.5.2  Vertex-specific network metrics A set of network metrics are similar to the geographic concepts of latitude and longitude, coordinates that identify each individual's position within a network. Paramount among these is the set of “centrality” measures, which describe how a particular vertex can be said to be in the “middle” of a network. In the 1970s and 1980s, the sociologist Phillip Bonacich developed a refined measure of centrality that took into consideration the different value a highly-connected person can have in contrast to people with a few rare connections. Network theorists noted that simply having many connections, called “degree centrality,” was only one way to be “at the center” of things. A person with fewer connections might have more rare and potentially “important” connections than someone with more connections. One connection can be more important than another in different ways. Some are better because they bridge across otherwise separated portions of the network, whereas others are important because they connect to wellconnected people. The following centrality metrics provide quantifiable measures for these concepts (see Chapter 6 for more details). Degree centrality Degree centrality is a simple count of the total number of connections linked to a vertex. It can be thought of as a kind of popularity measure, but a crude one that does not recognize a difference between quantity and quality. Degree centrality does not differentiate between a link to the CEO of a big company and a link to its most recent trainee hire. Degree is the measure of the total number of edges connected to a particular vertex. For directed networks where relationships have an origin and a destination rather than have mutual connections, there are two measures of degree: in-degree and out-degree. Indegree is the number of connections that point inward at a vertex. Out-degree is the number of connections that originate at a vertex and point outward to other vertices.

I.  Getting started with analyzing social media networks



3.5  Network analysis metrics

Betweenness centrality: Bridge scores for boundary spanners The notion of connection paths is central to the study of networks. Perhaps one of the most natural questions to ask about any two people in a network it is “How far apart are they?” This distance is measured simply: the distance between people who are not neighbors is measured by the smallest number of neighbor-to-neighbor hops from one to connect to the other. For instance, people who are not your neighbors, but are your neighbors' neighbors, are a distance 2 from you, and so on. The shortest path between two people is called the “geodesic distance” and is used in many centrality metrics. For example, betweenness centrality is a measure of how often a given vertex lies on the shortest path between two other vertices. This can be thought of as a kind of “bridge” score, a measure of how much removing a person would disrupt the connections between other people in the network. The idea of brokering is often captured in the measure of betweenness centrality. A “structural hole” is a term for recognizing a missing bridge. Wherever two or more groups fail to connect, one can argue that there is a structural hole, a missing gap waiting to be filled. Burt provides compelling evidence that individuals who bridge structural holes within their organizations are promoted faster than others [15]. Social network analysis has many strategic applications for people in an organization to analyze their position and the position of others. Managers and leaders can recognize gaps or disconnections within organizations and devote resources to bridging the divide. People may be able to apply social network analysis to identify locations in which a gap exists and elect to fill them, recognizing the value they can generate as broker between two otherwise separate groups. Closeness centrality: Distance scores for strategically located people Closeness centrality measures each individual’s position in the network via a different perspective from the other network metrics, capturing the average distance between each vertex and every other vertex in the network. Assuming that vertices can only pass messages to or influence their existing connections, a low closeness centrality means that a person is directly connected or “just a hop away” from most others in the network. In contrast, vertices in very peripheral locations may have high closeness centrality scores, indicating the high number of hops or connections they need to take to connect to distant others in the network. Think of closeness, paradoxically, as a “distance” score. Some people are just a few miles from the big city, others must drive for hours: similarly, people with high “closeness” centrality scores have many miles or rather personal connections that they must travel to reach many other people in the network. Note that in

41

some cases the inverse of the average distance to others in the network is used as a measure of closeness centrality. In that case, higher values indicate a more central position. Eigenvector and PageRank centrality: Influence scores for strategically connected people Eigenvector centrality is a more sophisticated view of centrality: a person with few connections could have a very high eigenvector centrality if those few connections were to very well-connected others. Eigenvector centrality allows for connections to have a variable value, so that connecting to some vertices has more benefit than connecting to others. The PageRank algorithm used by Google's search engine is a variant of Eigenvector Centrality, primarily used for directed networks. PageRank considers (1) the number of in-bound links (i.e., sites that link to your site), (2) the quality of the linkers (i.e., the PageRank of sites that link to your site), and (3) the link propensity of the linkers (i.e., the number of sites the linkers link to). See Chapter 6 for a more in-depth discussion and examples. Clustering coefficient: How connected are my friends? The clustering coefficient metric differs from measures of centrality. It is more akin to the density metric for whole networks, but focused on egocentric networks. Specifically, the clustering coefficient is a measure of the density of the 1.5-degree egocentric network for each vertex. When these connections are dense, the clustering coefficient is high. If your “friends” (alters) all know each other, you have a high clustering coefficient. If your “friends” (alters) don’t know each other, then you have a low clustering coefficient. People have different measures for their clustering coefficient depending on the ways they cultivate connections to others and the environments they are in.

3.5.3  Grouping, clustering, and community detection algorithms A network approach can discover and identify the boundaries of groups and clusters, or apply existing information about each vertex to create categories or divisions. In a network perspective, people maintain many relationships and are potentially members in many loosely defined groups and clusters. Defining exact group boundaries in a network may be difficult, reflecting the reality of people with multiple and shifting memberships. From a network perspective, a group is a collection of vertices. Groups can be formed for many reasons, in some cases some vertices are more connected to one another than they are to others. Relatively more cohesive or densely connected sets of vertices form regions, also called clusters, that may reflect the existence

I.  Getting started with analyzing social media networks

42

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

of groups. A group of people discovered in this way might not be explicitly named or recognized. Members of a network cluster might not recognize their collective membership despite their individual connections to others in the group. A rapidly growing body of research describes clustering algorithms, also called community detection algorithms, that automatically identify these clusters based on networks structures, as discussed in Chapter 7.

3.5.4  Structures, network motifs, and social roles Two people within a network may sometimes share a pattern of connection to other people, even if they do not connect to the same people. Certain professions have distinct patterns of connections, either linking with many others (real estate agents, and other retail professionals) or few (reclusive authors and artists, remote office workers, and some people whose work focuses on things rather than people). In addition to having the same the number of connections, some people share the same pattern of connections among the people with whom they connect. In some cases people are connected to people who are strangers to one another, in other cases a group may be densely connected to one another. These secondary patterns of connection are a distinctive feature of network analysis approaches: networks are as much about the attributes and patterns of connection among neigh-

bors as they are about the attributes and connections of any individual. Social roles are complex cultural and structural features of social life. An example social role like “father” is explicitly recognized in society, has a wide set of culturally shared meanings and expectations, is associated with particular goals and interests, and is partly defined by the content and structure of actions directed toward other distinctive role holders. Other types of social roles may not be as clearly defined or explicitly recognized by all the actors in a given social setting, but they have identifiable content, behavioral, and structural features. Studies of social media have illustrated the ways contributors create distinctive network patterns that reflect their role or status within the community (e.g., Welser, Gleave, and Smith [16]). These patterns are evidence of specialization of behavior in these social spaces. An example of a role in a social media space is the “answer person” who disproportionately provides the answers to questions asked in message board environments (see Chapter  10), “discussion people” who engage in extended exchanges of messages in large and populous threaded discussions, “discussion starters” who demonstrate influence over the topics discussed by the “discussion people,” “influential” people who are well connected to others who are more highly connected than they are, and boundary spanners who bridge between unconnected subgroups.

A D VA N C E D T O P I C

A renaissance of network research and data Since the 1960s, network analysis has blossomed. New research and methods have flourished and social networking has developed a new prominence in mainstream culture. Despite early challenges, in the past several decades a healthy and growing subfield has reemerged around social network analysis. New network tools and concepts have been created and applied to a wide and growing range of domains. Mathematical sociology has developed as a major subdiscipline in the social sciences, dedicated to finding elegant descriptions of complex social phenomena. Starting 80 years ago with simple handdrawn charts and diagrams that described small groups of people and their connections, network science concepts, methods, and tools are used today to calculate a range of measures that describe the shape, structure, and dynamics of potentially multi million or billion vertex networks. New methods have been developed for automatically organizing and displaying visualizations of the links among

large populations. This combination of structural models, visualizations, and metrics forms the key features of modern social network analysis. In the late 1960s, Stanley Milgram explored the idea of small world networks in a study that came to be referred to as “Six Degrees of Separation” [17], which later inspired the 1990 John Guare play and 1993 movie of the same name. The Milgram study explored the question of how connected any two people selected at random might be. Milgram sent a collection of letters to randomly selected people around the United States asking them to send the message to someone they knew who could move their letter closer to the target, a stock broker in Massachusetts. On average, the letters took six steps to arrive at their destination. The “six degrees” or steps suggested that even in large networks where most people are not directly connected, people can be reached from almost every other person through a small number of steps (although possi-

I.  Getting started with analyzing social media networks



43

3.5  Network analysis metrics

bly more than six, which was the average number of hops not the maximum!). Sampson’s study in the late 1960s of relationships among members of a residential monastery captured social network data during an event in which several members were expelled or chose to leave [18]. A series of social network datasets were collected by asking participants about who they liked and spent time with. Social network analysis of this data allowed Sampson to identify the future lines of division among the members of the network. The idea that members of a network can be grouped based on how densely they are connected is an important concept in network analysis. The sub-groups identified by network analysis can reflect important real world social divisions with consequences for the future of that network. For example, a notable study by Zachary in the 1970s mapped the structure of a Karate club based on affinities and connections between students and teachers. These maps predicted the ways the club eventually split when a new teacher, in conflict with the owner, left the studio and took many students with him [19]. The sociologist Barry Wellman demonstrated in the 1970s that real-world communities are composed of interlocking social networks of specialized relationships that changed dramatically in composition over a period of years. He proposed that society was now characterized by networked individualism in contrast to the clearly defined group memberships and identities of prior periods. Rather than defining oneself in professional or political terms, Wellman observed that people create personal networks in which they occupy distinct locations and roles. He later applied these techniques to study online networks [20]. In 1977, Wellman founded a social network analysis professional association, the International Network for Social Network Analysis (INSNA). INSNA now has more than a thousand members, many of whom have gathered for more than 20 years for an annual conference (“Sunbelt”) on social network analysis research.3 Journals and publications devoted to social network analysis include Social Networks, Connections, and the Journal of Social Structure. Social network data, methods, and visualizations appear across a much wider spectrum of journals and conference publications. In the early 1970s, the sociologist Mark Granovetter investigated the employment market, looking at how people discovered new job opportunities. He observed that, in contrast to the view held by classical economics, people were not freely floating independent actors in the labor market. They were embedded in a set of different relationships with particular people. Granovetter found that job news passed through connections that were not the closest and most intense relationships [21]. A person’s “weak

ties” brought news from distant parts of the social network to which “strong ties” did not have access because they occupied such a similar place in the network as the job seeker. Thus weak ties proved particularly useful for finding novel information, such as information about job prospects. Because weak ties were less intense, they were also less costly to maintain in terms of time and attention. As a result, it is possible to have many weak ties but only a few strong ties. Empowered with new network metrics and the means to calculate them, network analysts have focused on a variety of data sources and questions. Social networks have been applied to historical studies using records of investments, marriages, and memberships in elected positions. In the 1400s in the city of Florence, the Medici and Strozzi families struggled for domination. These families, along with many others, were locked in political struggles. In the 1970s, John Padgett collected records of the social relations among Renaissance Florentine families that he extracted from historical documents. Families were often connected through a variety of ties, relations, and business connections. A dataset was created that represented the financial loans, credits and joint partnerships, and marriages that bound families to one another. The resulting dataset included information about each family as well as their links to others. Each family had a value representing its net wealth in the year 1427, the number of seats it held in the local government between the years 1282 and 1344, and the number of business or marriage ties among the population of 116 families. Analyzing these data, Padget found that the Medici held great power because, he argued, they sat at the center of business and family networks, brokering connections that no other family could equal [22, 23]. A more modern version of the study of historical Florentine politics can be found in the study of interlocking directorships in modern corporations. Many corporations and other institutions have a board of directors, some of whom serve on more than one board. When board members serve on two or more boards, they link those corporations and, in aggregate, create interlocking directorships that combine to form even larger meta-institutions. By building on research on interlocking directorships in U.S. corporations [24, 25], websites like “They Rule” provide an interactive map that displays the common links between major corporations.4 In 1992, Robin Dunbar famously argued that people have an innate ability to handle a number of social relationships but not an endless number of them. Remembering people's names may have a biological limit as our brains evolved over long periods in which there were rarely more

3 www.insna.org.

4 www.theyrule.net.

Continued

I.  Getting started with analyzing social media networks

44

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

A D VA N C E D T O P I C ( c o n t ’ d ) than a few hundred people within any region, group, or tribe. The number 150 has been loosely associated with the idea of a “Dunbar” number, an upper limit on the number of relationships a person can normally manage.5 Other social animals with smaller brains have lower Dunbar numbers than humans, suggesting that complex social relationships require more mental resources (and the cranial volume to hold it). The Dunbar number advantage humans already have can be expanded with augmentation, through analog technologies like diaries, address books, the “filo-fax,” and now Friends and Contact list managers in social media platforms. Social media tools like Facebook, LinkedIn, text messengers and email contact lists extend our ability to maintain more relationships. These additional relationships can be said to be “weaker” than the core 150 “organic” relationships, but as Granovetter has shown, weak ties can collectively be of enormous strength and value.

Business applications of social network analysis Social network analysis has historically been an academic endeavor, but as network analysis tools and datasets become more available, pioneering businesses are applying it to help manage business challenges, gain insight into markets and communities, and build more robust industry relationships. For example, the work of Rob Cross and the Network Roundtable focuses on several practical applications of social network analysis for corporations and other large organizations, highlighting differences between healthy and underperforming divisions and the value of or-

5 www.lifewithalacrity.com/2004/03/the_dunbar_numb.html.

3.6  Social networks in the era of abundant computation The widespread adoption of networked communication technologies has significantly expanded the population of people who are both aware of network concepts and interested in network data. Although the idea of networks of connections of people spanning societies and nations was once esoteric, today many people actively manage an explicit social network of friends, contacts, buddies, associates, and addresses that compose their family, social, professional, and civic lives. Facebook posts forwarded from person to person have become a common and visible example of the ways information passes through networks of ­ connected

ganization spanning connections [26, 27]. Others apply network analysis to the improvement of corporate structures and processes [28]. In the early 1990s, Monge and Contractor [29] documented the many forms of social network patterns that emerge inside of organizations and institutions. Social networks have been shown to have a significant influence on the adoption of new technologies or social practices. The sociologist Everett Rogers described the concept of the “diffusion of innovations,” arguing that people with particular patterns of connections to others played pivotal roles in the success or failure of a new idea or message being rejected or adopted and distributed through the network [30]. Networks with different patterns of connection have different properties in terms of how they propagate a new message, rumor, or product and how they resist being dissolved when vertices are removed from the graph. These observations have significant implications for interventions into disease and rumor propagation and the cultivation of innovation [31]. Networks play an important role in e-commerce where collaborative filtering powers the familiar list of “books that people who liked this book also liked.” Businesses are also interested in learning the requirements of viral marketing. Diffusion can often lead to “cascades” where an unknown, even marginal idea can spread rapidly throughout the entire network and become widely observed, if still rare.6 Memes are a commonly-cited example of contagion, as are viral messages, such as viral videos on YouTube that go from dozens to millions of viewers in a few months, weeks or even days. 6 https://5harad.com/papers/twiral.pdf; https://5harad.com/ papers/diffusion.pdf.

­ eople. The notion of “friends of friends” is now easy p to illustrate in the features of social media applications like Facebook and LinkedIn that provide explicitly named “social networking” services. Viral videos and chain emails illustrate the way word of mouth has moved into computer-mediated communication channels. The idea of “six degrees of separation” has moved from the offices of Harvard sociologists to become the dramatic premise of a Broadway play to now appear as an expected feature of services that allow people to browse and connect to their friend’s friends. As network concepts have entered everyday life, the previously less visible ties and connections that have always woven people together into relationships, cliques, clusters, groups, teams, partnerships, clans,

I.  Getting started with analyzing social media networks



45

3.6  Social networks in the era of abundant computation

tribes, coalitions, companies, institutions, organizations, nations, and populations have become more apparent. Patterns of information sharing, investment, personal time and attention have always generated network structures, but only recently have these linkages been made plainly visible to a broad population. In the past few decades, the network approach to thinking about the world has expanded beyond the core population of researchers to a wide range of analysts and practitioners who have applied social network methods and perspectives to understand their businesses, communities, markets, and disciplines. Today, because many of us manage many aspects of our social relationships through a computer-networked social world, it is useful for many more people to develop a language and literacy in the ways networks can be described,

analyzed, and visualized. Visualizing and analyzing a social network is an increasingly common personal or business interest. The science of networks is a growing topic of interest and attention, with a growing number of courses for graduates and undergraduates, as well as educational materials for a wider audience (e.g., television documentaries).7 The availability of cheaper computing resources and network datasets has enabled a new generation of researchers access to studies of the structures of social relationships at vastly larger scale and detail. Since the late 1960s, as computing resources and network datasets have grown in availability and dropped in cost, researchers began developing tools and concepts that enabled a wider and more sophisticated application of social network analysis.

A D VA N C E D T O P I C

Social network analysis research meets the web As access to electronic networks grew in the 1970s, academic and professional discussions and collaborations began to take place through them. Systems to support the exchange of messages and the growth of discussions and even decisions became a major focus of systems development and the focus of study itself. Freeman and Freeman [32] collected data from the records of the Electronic Information Exchange System (EIES) that itself hosted a discussion among social network researchers. Two relations were recorded: the number of messages sent and acquaintanceship. These systems became the focus of the first systematic research into naturally occurring social media. Even before the Internet, early computer network applications supported the creation of exchanges, discussions, and therefore social networks, built by reply connections among authors. Early proprietary systems evolved into the public World Wide Web. In the 1990s, the computer scientist Jon Kleinberg created an algorithm called HITS that identified the patterns of links between high-quality web pages. This algorithm later inspired Stanford graduate students Sergey Brin and Larry Page who founded the Google corporation to develop a further refinement they called “Page Rank.” Kleinberg’s work described different locations within a population of linked documents on the World Wide Web: not all documents are equal. On the Web, a document or page can link to another page, forming a complex network

of related documents. Some documents contain many pointers to other documents, whereas others have many documents that point at them. These “hubs” and “authorities” defined two broad classes of web pages that offered a path to identifying high-quality content. Links from one page to another are considered to be indicators of value. Refinements of the HITS algorithm made use of eigenvector centralities to implement the page rank algorithm that is the core of the Google “Page Rank” web search ranking method [33]. Network researchers studying social networks and the Internet found that empirical networks often exhibit “small-world properties”: most nodes are not neighbors with each other, but most nodes can be reached from almost every other node in a small number of hops. In the late 1990s, the physicist/sociologist Duncan Watts, working with the mathematician Steven Strogatz, created mathematical models of “small world” networks and contrasted them with purely random networks such as those proposed by Erdos and Renyi [34]. Their model captured the natural properties of social networks far better than those that assumed a purely random or normal distribution of links. Although most people have connections to other people who are local to them, people occasionally have a few connections that link them to another person physically far from the individual. Many of our friends are likely to live or work near us, but a few may be very Continued

7 “Connected: The Power of Six Degrees,” http://ivl.slis.indiana.edu/ km/movies/2008-talas-connected.mov.

I.  Getting started with analyzing social media networks

46

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

A D VA N C E D T O P I C ( c o n t ’ d ) far away. Even a modest number of these relatively rare far-reaching links can dramatically change the properties of a network, making the widespread transmission of messages much easier. This model significantly improved on earlier ways of thinking about network growth and structure, better approximating the observed structure of naturally occurring social networks. Later researchers have built upon their work to devise models that generate “small world” networks that more closely match empirical networks, helping us to understand how networks may have become the way they are. For example, Barabasi and Albert have developed a family of models of preferential attachment that can generate “scale-free” networks, which are a common feature of social networks [35]. Scale free networks have a power law degree distribution, meaning that there are a few key hubs in a network and many poorly connected vertices. While none of these models perfectly predict real world observed social networks, they provide a method for systematically comparing networks and focus attention on the processes that may have led to the characteristics that we do see in the networks around us. In the past few years, researchers have begun to study large web-based networks. For example, Leskovec and Horvitz calculated metrics for a graph that includes more than 300 million users of the Microsoft Messenger service [36]. Each user typically had one or more “buddies” with whom he or she might send one or more messages and receive some in return. Buddies often listed their locations, allowing these linkages to be aggregated into a complex map of the world and the flow of conversation around it. Others have reported on the hyperlink network created by web pages hyperlinking to other web pages (e.g., Park and Thelwall [37]). A number of studies have examined the blog network. For example, Adamic and Adar [38] showed how political blogs are divided into two clear clusters with minimal overlap that represent the left and right political populations in the United States. More recently, Kelly and Etling mapped Iran’s blogosphere, identifying more than 20 subcommunities of bloggers who wrote in Farsi for an Iranian audience.8 Another line of research has focused on visualizing social networks. For example, an early influential paper by Heer and Boyd [39] described a tool called Vizster that allowed users to navigate through their friends from a social networking site to explore social connections. Now there are entire conferences dedicated to network visualization, 8 http://cyber.law.harvard.edu/publications/2008/ Mapping_Irans_Online_Public.

such as the annual Graph Drawing and Network Visualization symposium. As social media has matured and its ability to shape perceptions has been recognized, it has become the target of concerted efforts at manipulation. Misinformation, disinformation and propaganda have grown in visibility and concern. Claims and counter claims of “fake news” have become common. While initial critical analysis of social media like the work of Eli Parisier focused on its divisive ability to create “filter bubbles,” later work has explored the ways the bubbles can be penetrated and the divisions between them amplified.9 Computer scientist Kate Starbird shows the ways that national, political and commercial groups have collaborated to influence social media by creating messages aimed at making already divided groups more extreme.10 Fil Menczer and collaborators are building tools to identify “bots” and address the propagation of disputed or low value information.11 Information networks are designed to move all kinds of information, not just the true or high value kinds. Paradoxically people often resist abandoning beliefs despite strong evidence and often increase their commitments to beliefs that are challenged. Since a large amount of the information people want to create and consume is explicitly “fiction,” the goal of building machines that can highlight facts and diminish fake information is a challenge. The concept of “tribal epistemology” suggests that most facts exist only for certain people in certain places and times. If multiple truths can coexist it may be better to build maps of the many beliefs and their believers. Rather than seeking to identify the true among the field of untrue material, an alternative approach could map the range of claims and the people and groups who make them. Social media is often thought of as an example of a “marketplace of ideas”.12 If so, social media network analysis could be thought of as a form of accounting software for this marketplace. Without accounting and auditing most markets become rife with fraud and marketplaces for ideas are no exception. With better accounting it should be possible to clearly trace which groups are the source and support for various ideas, claims and beliefs. When independent scrutiny of markets is possible manipulation and collusion can be identified and potentially addressed. 9 https://en.wikipedia.org/wiki/Filter_bubble. 10 https://medium.com/s/story/the-trolls-within-how-russianinformation-operations-infiltrated-online-communities691fb969b9e4. 11 http://cnets.indiana.edu/blog/2018/03/08/ science-fake-news/. 12 https://en.wikipedia.org/wiki/Marketplace_of_ideas.

I.  Getting started with analyzing social media networks



47

3.8  Tools for network analysis

3.7  The era of abundant social networks: From the desktop to your hand We now live in a new era of network data abundance. Network data collection was once a time-consuming and laborious process that yielded small datasets at great cost. Observations, surveys and interviews took many days or weeks to perform, could not be repeated frequently, required many people to produce, and often yielded low rates of participation with inherent biases and errors. Asking people about their relationships with others continues to have benefits and offers unique sources of insight, but people have been shown to be a poor source of accurate information as bias and faulty memory warp what people report about who they know and with whom they interact. The challenge of creating a dataset that spanned long periods or large numbers of people or contained records of many events proved insurmountable using traditional methods. Today, interactions between people increasingly take place through computing systems. Users create many types of networks in a machine-readable form each day as our interactions are documented in a computer. When we use these communication tools, databases are created and maintained with records and log files that document the details of the time, place, and participants of each interaction, whether via computers or telephones or even televisions. These event logs describe many different kinds of connection but share a common structure in which one person or entity is linked to another by some relationship. The creation of these machine-readable network datasets mean that long periods of time or large populations connected by many events can now be studied using widely available computing equipment and data sources. Like a jump from Galileo’s handmade telescope to the orbiting Hubble, network science has made a vast leap in scale and scope as we create a digitally networked world around ourselves. The historical drought of social network data has ended with a flood of new sources of network data. The challenge has shifted to rapidly develop tools and concepts needed to process and analyze this deluge of connected data. Technical methods for building multi-terabyte databases have shifted to the even vaster task of managing petabytes of data. New methods of harnessing thousands and even millions of computers in parallel have been driven by the growing need to manage vast data stores growing from the web. The challenge is likely to grow steeper as new sources of network data come pouring out off an emerging class of sensor-rich devices (the “Internet of Things”) that record vast streams of data from billions of people, devices, and locations. The early wave of this surge of data can be seen in new sources of data from everyday life that are

being captured and recorded with mobile and wearable devices, creating a new stream of archival material that is richer than all but the most obsessively observed biographies. It has become common in recent years that the most timely and well-placed photographs and video recordings have come from everyday individuals with phones and computers rather than from news photographers and reporters. The coming wave of mobile technologies is likely to deepen this trend, with new ways for smartphones or other devices to capture information about their users and the relationships and world around them. Many mobile applications integrate location into their service (see Chapter  2). As phones are aware of their location, a new set of mobile social software applications are becoming possible, as evidenced by new services such as Strava, a good example of a mobile data collection, analysis, and presentation service for cyclists, runners, and other trail sports. Other products like FitBit13 and Apple Watch are examples of social location and vital signs recording technologies that enabled web applications to provide self-monitoring medical and fitness tracking. Medical communities overlap with trail-based exercise communities by using devices that extensively quantify your “self” and “others.” These devices enable consumers to collect detailed medical readings nearly all the time that are cross referenced by location. The result is a growing aggregated map of the health and environmental conditions of the planet, not unlike early examples of collectively authored road maps of whole nations accomplished by the Open Street Map project.14

3.8  Tools for network analysis The growth of interest in network analysis has been dramatic, but until recently the development of social network analysis tools has lagged, and they remained challenging for many non-technical people to use. Applying network approaches has been traditionally a challenge that involved much more than simply mastering a new set of concepts and ideas that focus on relationships and patterns. Network data have traditionally been difficult to create and collect, and the tools for analyzing and visualizing networks have demanded significant technical skill and often mastery of programming languages. Many tools that exist to support network analysis demand significant commitment to learn and master. The existing network tools that are relatively easier to use have typically lacked support for easily importing social media network data. In the past few years, many network analysis projects and research papers have focused 13 www.fitbit.com. 14 www.openstreetmap.org/.

I.  Getting started with analyzing social media networks

48

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

on computer-mediated networks of people, documents, and systems. Only recently have new tools made it simpler for people to extract data from major social media network sources and to perform a basic network analysis workflow without requiring programming skills or using a command line interface. Social media network data collection, scrubbing, analysis, and display tasks have historically required a remarkable collection of tools and skills. While tools like Datasift make data available from numerous social media platforms, significant technical skills are needed to connect to application programmer interfaces (APIs). In contrast, this book focuses on a single tool designed for non-programmers, NodeXL, because of its relative ease of use, support for rich visuals and analytics, and integration with the ubiquitous Excel spreadsheet software. The python or “R” programming language path is certainly the high road for experts and those with demanding volumes of data or esoteric data analysis requirements. But for the noncoding user, NodeXL may be one of the easiest ways to both manipulate network graphs and get graph datasets from a variety of social media sources. A detailed step-by-step guide to the core features of NodeXL can be found in Part II of the book.

3.9  Node-link diagrams: Visually mapping social networks One of the key elements that characterizes modern social network analysis is the use of visualizations of complex networks. Compared to staring at edge lists or network matrices (see Section  3.2.4), looking at a network graph can provide an intuitive visual overview of the structure of the network, calling out cliques, clusters, communities, and key participants. It could be said that a graph visualization is worth a thousand ties. Not only can network visualizations inspire understanding and insights, they can also be appealing and even beautiful. They can serve as persuasive tools that demonstrate important points about networks. The ability to map attribute data and network metric scores to visual properties of the vertices and edges (see Chapters  5 and 6) makes them particularly powerful. However, network visualizations are often as frustrating as they are appealing. Network graphs can rapidly get too dense and large to make out any meaningful patterns as illustrated in Figure 3.5. Many obstacles like vertex occlusions and edge crossings make creating well-organized and readable network graphs challenging. There is an upper limit on the numbers of vertices and edges that can be displayed in a bounded set of pixels; typically only a few hundred or thousand vertices can be meaningfully and distinctly represented on average-sized computer screens.

In his appeal for better-quality network visualization, Shneiderman [40] has suggested that we aspire to reach the worthy but not always attainable goal of “netviz nirvana” in which the following goals are proposed: • Every vertex is visible. • Every vertex’s degree is countable (i.e., the number of connections that start or end at that vertex). • Every edge can be followed from source to destination. • Clusters and outliers are identifiable. To approach netviz nirvana, careful preparation, layout, and filtering techniques must be used. In practice, network visualizations often fall far from the mark. However, the graphs shown throughout this book illustrate the value of carefully crafting network graphs. We hope they will inspire network analysts to take the care needed to create substantive, understandable, and esthetically pleasing graphs.

3.10  Common network analysis questions applied to social media Once a set of social media networks has been constructed and social network measurements have been calculated, the resulting dataset can be used for many applications. For example, network datasets can be used to create reports about community health, comparisons of subgroups, and identification of important individuals, as well as in applications that rank, sort, compare, and search for content and experts. The value of a social network approach is the ability to ask and answer questions that are not available to other methods. Network methods focus on the patterns of relationships in contrast to the volumes of individuals. Although analysts, marketers, and administrators often track social media participation statistics, they rarely consider measures of network position and structure. Traditional participation statistics can provide important insights into the volume of engagement of a community, but can say little about the structure of the connections between community members. Network analysis can help explain important social phenomena such as group formation, group cohesion, social roles, personal influence, and overall community health. Combining traditional participation metrics with network metrics provides the best of both worlds and allows you to answer important questions such as the following: • What kinds of social roles are being performed within a social media collection? Does a community have enough people filling the important roles? • Which individuals play important social roles within a group or collection? Who would make a good administrator based on that person’s network position?

I.  Getting started with analyzing social media networks



3.11  Practitioner’s summary

49

FIGURE  3.5  A medium-sized node-link network diagram visualization of Twitter users linked by patterns of following. This sized graph illustrates many issues with a network graph containing more than a few dozen vertices. Many vertices sit on or overlap with other vertices. The number of edges associated with some vertices is impossible to count, whereas other edges cannot be traced from source to destination. Improvements to network layout are an active area of research.

• What subgroups exist? Do connections between subgroups exist? Who plays the bridge roles that connect otherwise unconnected groups? • How do new ideas propagate through a network? Who are the influencers that spark the spread of ideas? • How do the overall structures of a social network change after a particular event (e.g., a company social, a round of new hires or layoffs, a product launch or recall)?

3.11  Practitioner’s summary The opportunities for practitioners to apply network analysis to contemporary business, community management, political influence, and team collaboration have dramatically increased in recent years. The once esoteric concepts and metrics of network analysis have become talk show and airport lounge topics. The difficulties in collecting and analyzing network data have been dra-

matically reduced by powerful database methods and well-designed network analysis and visualization tools. There is still a lot of work to be done, but practitioners now have the potential to make more effective decisions based on network analyses of their own data conducted in a few hours, rather than a few months. Learning network concepts and tools is a necessary first step, but the payoffs for applying network methods are large. The growing numbers of trained social media network analysts and consultants are complemented by a vast array of books and informative websites, online seminars, and Wikipedia pages which make the necessary training widely available. At the same time, network analysis methods are rapidly spreading through university curricula and filtering into high school courses. Attending public seminars and professional conferences provides other means to acquire skills and make valuable connections. Your first steps may be a struggle, but we hope that with each step the processes become smoother and the professional benefits larger.

I.  Getting started with analyzing social media networks

50

3.  Social network analysis: Measuring, mapping, and modeling collections of connections

3.12  Researcher’s agenda The research progress on network analysis has been dramatic in the past few decades, transforming an exotic research topic into a thriving research community in academia, government, and industry. The existing metrics, clustering, and layout algorithms are stabilizing, but innovative approaches are still emerging to trigger bursts of new research. As practitioner pressure builds to apply network analysis to ever larger datasets, researchers have developed remarkably more efficient algorithms, while hardware developers have produced powerful graphics processors (based on gaming computers), huge arrays of computers, and scalable cloud computing services. Meanwhile, new social media services generate more relational data than ever before, ushering in a golden era of social science research on human relationships and collaboration. The algorithms and hardware provide the platforms, but the concomitant development of vastly improved user interfaces for network analysis has begun to enlarge the community of users from the dedicated sociologists who are also programmers to the broad segment of business analysts who use spreadsheets or simplified web-based tools. Packaging the complex processes of frequently applied network analyses into a few clicks is the next challenge in many fields, thereby inspiring other researchers and developers to simplify the processes even further, while increasing the power offered to users. The best is yet to come.

References [1] J.P.  Scott, Analysis, Social Network, fourth ed., Sage, Thousand Oaks, CA, 2017. [2] A.L.  Barabási, Network Science, Cambridge University Press, Cambridge, UK, 2016. [3] M.  Newman, Networks, second ed., Oxford University Press, New York, NY, 2018. [4] C. Kadushin, Understanding Social Networks: Theories, Concepts, and Findings, Oxford University Press, New York, NY, 2012. [5] S.P.  Borgatti, M.G.  Everett, J.C.  Johnson, Analyzing Social Networks, second ed., Sage, Thousand Oaks, CA, 2018. [6] S.  Wasserman, K.  Faust, Social Network Analysis: Methods and Applications (Structural Analysis in the Social Sciences), Cambridge University Press, Cambridge, 1994. [7] B.  Wellman, Structural analysis, in: B.  Wellman, S.D.  Berkowitz (Eds.), Social Structures, Cambridge University Press, Cambridge, 1988, pp. 19–61. [8] J.L. Moreno, Who Shall Survive? A New Approach to the Problem of Human Interrelations, Nervous and Mental Disease Publishing Co., Washington, 1934. [9] A.  Davis, B.B.  Gardner, M.R.  Gardner, Deep South: A social Anthropological Study of Caste and Class, University of Chicago Press, Chicago, Ill, 1941. [10] L.C.  Freeman, The Development of Social Network Analysis: A Study in the Sociology of Science, BookSurge, LLC, North Charleston, SC, 2004. [11] F.  Roethlisberger, W.  Dickson, Management and the Worker, Cambridge University Press, Cambridge, UK, 1939. [12] G. Homans, The Human Group, Harcourt-Brace, New York, 1950.

[13] R. Breiger, S. Boorman, P. Arabie, An algorithm for clustering relational data with applications to social network analysis and comparison with multidimensional scaling, J Math Psychol 12 (1975) 328–383. [14] S.F. Nadel, The Theory of Social Structure, Cohen & West, London, 1957. [15] R. Burt, Brokerage and Closure: An Introduction to Social Capital, Oxford University Press, Oxford, 2005. [16] H. Welser, E. Gleave, M. Smith, Visualizing the signatures of social roles in online discussion groups, J. Soc. Struct. 8 (2) (2007). [17] S.  Milgram, The small world problem, Psychol. Today 2 (1967) 60–67. [18] S. Sampson, Unpublished doctoral dissertation, Cornell University, in: Crisis in a cloister, 1969. [19] W. Zachary, An information flow model for conflict and fission in small groups, J. Anthropol. Res. 33 (1977) 452–473. [20] B. Wellman, An electronic group is virtually a social network, in: K. Sara (Ed.), Culture of the Internet, Lawrence Erlbaum, Mahwah, NJ, 1997. [21] M. Granovetter, The strength of weak ties, Am J Sociol 78 (6) (1973) 1360–1380. [22] J.  Padgett, C.  Ansell, Robust action and the rise of the medici, 1400–1434, Am. J. Sociol. 98 (6) (1993) 1259–1319. [23] D.  Kent, The Rise of the Medici: Faction in Florence, 1426–1434, Oxford University Press, Oxford, 1978. [24] M. Mizruchi, L.B. Stearns, A longitudinal study of the formation of interlocking directorates, Adm. Sci. Q 33 (2) (1988) 194–210. [25] B. Mintz, M. Schwartz, The Power Structure of American Business, University of Chicago Press, Chicago, 1985. [26] R. Cross, R.J. Thomas, Driving Results through Social Networks: How Top Organizations Leverage Networks for Performance and Growth, Jossey-Bass, San Francisco, CA, 2009. [27] R. Cross, R.J. Thomas, Driving Results through Social Networks: How Top Organizations Leverage Networks for Performance and Growth, John Wiley & Sons, San Francisco, CA, 2009. [28] M.  Kilduff, W.  Tsai, Social Networks and Organizations, Sage, Thousand Oaks, CA, 2003. [29] P.R. Monge, N. Contractor, Theories of Communication Networks, Oxford University Press, New York, 2003. [30] D.E.M.  Rogers, Diffusion of Innovations, fifth ed., Simon and Schuster, New York, 2003. [31] M.E.J. Newman, The structure and function of complex networks, SIAM Rev. 45 (2003) 167–256. [32] L.C. Freeman, Centrality in social networks: conceptual clarification, Social Netw. 1 (1979) 35–41. [33]. S. Brin, L. Page, The anatomy of a large-scale hypertextual web search engine, in: Proc. 7th World-Wide Web Conference (WWW7), BrisBane, Australia, 1998. [34] D.J. Watts, S.H. Strogatz, Collective dynamics of ‘small-world’ networks, Nature 393 (6684) (1998) 440–442. [35] A.L.  Barabasi, R.  Albert, Emergence of scaling in random networks, Science 286 (1999) 509–512. [36] J.  Leskovec, E.  Horvitz, Planetary-scale views on a large ­instant-messaging network, in: Proc. 17th International Conference on World Wide Web (Beijing, China, April 21–25, 2008). WWW ’08. ACM, New York, NY, 2008, pp. 915–924. [37] H.W.  Park, M.  Thelwall, Hyperlink analyses of the world wide web: a review, J. Comput. Mediated Commun. 8 (4) (2003). [38] L.A. Adamic, E. Adar, Friends and neighbors on the web, Social Netw. 25 (3) (2003) 211–230. [39] J. Heer, D. Boyd, Vizster: visualizing online social networks, in: Proc. 2005 IEEE Symposium on Information Visualization (October 23–25 2005), INFOVIS. IEEE Computer Society, Washington, DC, 2005. [40] B.  Shneiderman, A.  Aris, Network visualization with semantic substrates, IEEE Trans, Visualization Comput. Graphics 12 (5) (2006) 733–740.

I.  Getting started with analyzing social media networks



Additional resources

Additional resources A.L.  Barabasi, Linked: How Everything is Connected to Everything Else and What It Means, Penguin Group, New York, 2003. S. Borgatti, A. Mehra, D. Brass, G. Labianca, Network analysis in the social sciences, Science 323 (5916) (2009) 892–895. M. Buchanan, Nexus: Small Worlds and the Groundbreaking Theory of Networks, Norton, New York, NY, 2002. R. Burt, Structural Holes: The Social Structure of Competition, Harvard University Press, Cambridge, MA, 1995. R.S. Burt, Structural Holes, Harvard University Press, Cambridge, MA, 1992. W.D. Nooy, A. Mrvar, V. Batagelj, Exploratory Social Network Analysis with Pajek: Revised and Expanded Edition for Updated Software, Third Edition, Cambridge University Press, Cambridge, UK, 2018.

51

S.  Johnson, Emergence: The Connected Lives of Ants, Brains, Cities, and Software, Penguin, London, UK, 2002. W. Nooy, M. De, A., & Batagelj, V., Exploratory Social Network Analysis With Pajek, Cambridge University Press, Cambridge, UK, 2005. W.D. Nooy, A. Mrvar, V. Batagelj, Exploratory Social Network Analysis with Pajek: Revised and Expanded Edition for Updated Software, third ed., Cambridge University Press, Cambridge, UK, 2018. M.A.  Russell, M.  Klassen, Mining the Social Web: Data Mining Facebook, Twitter, LinkedIn, Instagram, GitHub, and More, O'Reilly Media, Inc. Sebastopol, CA, 2019. D.  Watts, Small Worlds, Princeton University Press, Princeton, NJ, 1999. D. Watts, Six Degrees, Norton, New York, 2003. B. Wellman, S.D. Berkowitz, Social Structures: A Network Approach, Cambridge University Press, Cambridge, UK, 1988.

I.  Getting started with analyzing social media networks

P A R T

I I

NodeXL tutorial: Learning by doing This section (Chapters  4 through 8) introduces the NodeXL tool for analyzing networks using a hands-on approach. It assumes no knowledge of network analysis tools and assumes only a basic familiarity with social network analysis concepts (see Chapter  3). The section has been used as the basis for workshops and classroom lab sessions and takes approximately 8 h to complete with participants following along, though

a­ bbreviated versions can also work. It also serves well as a ­walk-through of NodeXL, introducing all of the basic functionality. We recommend working through the examples using the datasets available at the book’s accompanying website:  https://www.smrfoundation.org/ nodexl/teaching-with-nodexl/teaching-resources/​. The examples found in the chapters in Part III assume you have read the Part II chapters.

C H A P T E R

4 Installation, orientation, and layout O U T L I N E 4.1 Introduction

55

4.2 Downloading and installing NodeXL

56

4.3 Getting started with NodeXL 4.3.1 Opening a new NodeXL file 4.3.2 NodeXL menu ribbon 4.3.3 Spreadsheet and graph pane 4.3.4 Manually entering data 4.3.5 Importing data 4.3.6 Showing the graph 4.3.7 Highlighting an edge or vertex 4.3.8 Resizing and moving the graph pane

56 56 56 56 56 56 58 58 58

4.4 Layout: Arranging vertices in the graph pane 4.4.1 Manual layout 4.4.2 Automatic layout 4.4.3 Adjusting Fruchterman-Reingold settings 4.4.4 Updating the graph pane 4.4.5 Preserving a layout 4.4.6 Graph pane tools

59 59 59 60 61 61 61

4.1 Introduction The NodeXL Template for Microsoft Excel is an add-on to the spreadsheet application that provides a range of basic network analysis and visualization features. It is maintained by the Social Media Research Foundation (https://www.smrfoundation.org), a non-profit organization dedicated to the provision of tools and research to aid in “mapping, measuring and understanding the landscape of social media.” NodeXL’s central goal is ease of use, making it ideal for students and professionals to learn the concepts and methods of social network analysis with visualization as a key component. NodeXL supports the calculation of frequently used network metrics, grouping (i.e., clustering) of vertices, and textual analysis, though it does not support all of the advanced analysis techniques available in research-oriented tools such as Pajek, UCINet, or R. As a network visualization tool NodeXL is unparalleled, supporting a variety of network layouts, manual

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00004-2

4.4.7 Graph pane options

61

4.5 Undirected and directed graph type 4.5.1 Changing the type of network 4.5.2 Reciprocated edges

62 62 63

4.6 Working with NodeXL files 4.6.1 Saving NodeXL files 4.6.2 Exporting an existing NodeXL file 4.6.3 Opening an existing NodeXL file 4.6.4 Opening a NodeXL file created on another computer 4.6.5 Creating a trusted location for NodeXL files from the Internet

63 63 63 63

4.7 Practitioner’s summary

64

4.8 Researcher’s agenda

65

References

65

NodeXL papers

66

63 64

control of vertex placement, labeling of vertices, edges, and groups, and visual properties such as color, opacity, size, tooltips, images, etc. NodeXL is also unique in its ability to help analysts capture a range of social media network datasets without any programming, as illustrated in Part III of this book. NodeXL’s integration with Excel provides significant computational power, flexibility, and familiarity for many, though it is limited in its ability to analyze very large networks with hundreds of thousands of edges. It also boasts a highly active community of users who share files and images at the NodeXL Graph Gallery (https://nodexlgraphgallery. org), and communicate with the developers and one another on Twitter (@smr_foundation), Facebook,1 and a monthly newsletter.2

1

https://www.facebook.com/Social.Media.Research.Foundation/.

2

https://www.smrfoundation.org/newsletter/.

55

© 2020 Elsevier Inc. All rights reserved.

56

4.  Installation, orientation, and layout

NodeXL comes in two flavors. NodeXL Basic is free and allows users to create and visualize networks, as well as perform basic network and grouping (i.e., clustering) analyses. It is best used as a viewer for files created in NodeXL Pro, the full-featured, ­subscription-based version of NodeXL. NodeXL Pro supports the calculation of advanced metrics, content analysis, use of social media network importers, data exporting, and automation of tasks, among other things. Educational discounts are available for NodeXL Pro. This book covers the shared features, as well as features only associated with NodeXL Pro.

4.2  Downloading and installing NodeXL Detailed instructions on how to download both versions of NodeXL are found at https://www.smrfoundation.org/nodexl/installation/. NodeXL works with Excel 2007, 2010, 2013, and 2016 and is anticipated to work with future releases of Excel. It does not work with Mac versions of Excel. If you would like to run NodeXL on a Mac, you will need to boot into Windows using Boot Camp or another software hypervisor such as VMware Fusion. If it is the first time you are installing an Excel template, the setup program will install several prerequisites. Once installed, NodeXL will automatically update when opened. This means there may be discrepancies with the NodeXL version you are using and the one used in the book examples. However, most of the core features will remain the same. New features are the most common type of change and a full release history and explanations of upgrades are provided on the Social Media Research Foundation blog (https://www.smrfoundation.org/blog/).

4.3  Getting started with NodeXL 4.3.1  Opening a new NodeXL file Once NodeXL has been installed, open a new NodeXL file by clicking on the NodeXL Excel Template in the Start Menu. This will open a blank NodeXL file. NodeXL files end in the .xlsx extension just like Excel files, but when you open them, they will load the NodeXL add-on features like the graph display pane and custom menu items.

4.3.2  NodeXL menu ribbon Display the NodeXL menu ribbon (Figure  4.1) by clicking on the NodeXL tab in the Excel menu ribbon. The NodeXL menu provides access to all of the NodeXL features from a single place. It is organized into sections: Data, Graph, Visual Properties, Analysis, Options, Show/Hide, and Help. Hovering over certain buttons displays additional information about that feature. The Help control on the far right (Figure 4.1) describes each feature including those added after this book is published.

4.3.3  Spreadsheet and graph pane NodeXL is composed of two major sections: the spreadsheet where data is stored and the graph pane (called Document Actions) used to display network visualizations (Figure 4.2). The spreadsheet workbook includes multiple worksheets, each of which contains a different type of network data. For example, the Vertices worksheet contains a row for each vertex (e.g., each person in a network), while the Edges worksheet contains a row for each edge (e.g., each connection between 2 vertices). When NodeXL starts up, the graph pane shows a default splash screen (Figure 4.2) until a network is visualized.

4.3.4  Manually entering data One way to begin using NodeXL is to manually type in an edge list. Navigate to the Edges worksheet and enter in the names shown in Figure 4.3. Each row of the edge list represents a single edge between the two people specified in the Vertex1 and Vertex2 columns. The fictional data represents LinkedIn connections between a subset of employees of a large company we’ll call Analyzing Big Complex Data (ABCD for short). These connections are undirected; that is, the first row showing that Ben is connected to Ava also implies that Ava is connected to Ben.

4.3.5  Importing data NodeXL can import data from a variety of formats and data sources. To import data, choose the data source on the Import drop-down available in the Data section of the NodeXL Ribbon (Figure 4.1). The options at the top allow you to import data from other network a­ nalysis

FIGURE 4.1  NodeXL menu ribbon.

II.  NodeXL tutorial: learning by doing

FIGURE 4.2  NodeXL edges worksheet (left) and graph pane with splash screen (right).

FIGURE 4.3  ABCD company network edges worksheet (left) and default show graph results (right). The red edge is highlighted when a cell in the corresponding edge row is selected (e.g., the Camila and Ava edge).

II.  NodeXL tutorial: learning by doing

58

4.  Installation, orientation, and layout

programs such as UCINET and Pajek, as well as files from the generic GraphML network data format. You can also open data from another opened excel file that may include an edge list (From Open Workbook…) or an adjacency matrix (From Open Matrix Workbook…) (see Chapter  3 and [1]). A variety of built-in or third party data importers can be used to dynamically download data from social media sites such as Twitter and YouTube or from email collections (see Part III).

4.3.6  Showing the graph Next, click on the Show Graph button (directly above the graph pane) to visualize the network (Figure  4.2). This will change the text on the button to Refresh Graph (Figure  4.3). It will also automatically create a row on the Vertices worksheet for each of the 13 unique people (Figure 4.4). This operation can also be manually performed without displaying the graph visualization by selecting the NodeXL>Data>Prepare Data>Get Vertices from Edges Worksheet function. Use this approach when working with very large datasets that need to be filtered before they are visualized.

4.3.7  Highlighting an edge or vertex The data in the spreadsheet and the graph pane are coupled so that clicking on a row in the Edges worksheet will highlight the corresponding edge in the graph (Figure  4.3). Vertices can be highlighted by selecting a specific row in the Vertices worksheet (Figure 4.4). And visa versa, clicking on a specific vertex in the graph pane will automatically highlight the corresponding row in the spreadsheet. You can even click on multiple rows to highlight all related edges or vertices.

4.3.8  Resizing and moving the graph pane The graph pane can be resized by dragging the lefthand side of the pane to the left or right. It can also be undocked from the spreadsheet by clicking on the Document Actions title and dragging it outside of the Excel window. This is recommended when working with multiple monitors.

FIGURE 4.4  Vertices worksheet with ABCD network vertices and Ava selected. Graph laid out with the Harel-Koren Fast Multiscale algorithm as described in Section 4.4.2.

II.  NodeXL tutorial: learning by doing



4.4  Layout: Arranging vertices in the graph pane

4.4  Layout: Arranging vertices in the graph pane The location of vertices in a network visualization can make a significant difference in how easy it is to understand and gain insights into a network. For example, compare the visualizations of the ABCD network in Figures 4.3 and 4.4. These display the same network, but the layout in Figure  4.4 is far more readable. Creating readable graphs can be challenging, particularly for large networks. However, there are some useful heuristics to help improve your graphs (see Advanced topic: Readability network layout heuristics). Fortunately, NodeXL provides numerous techniques to manually and automatically adjust network layouts.

59

edges) if there are more than one. If there are isolates (i.e., individual vertices not connected to any others), visually separate them from the rest of the vertices. Do this automatically using the Layout Options dialog (Figure 4.6) by choosing Lay out the graph’s smaller connected components in boxes at the bottom of the graph pane. If there are clear sub-groups (i.e., clusters of highly connected vertices as discussed in Chapter 7), consider separating them out into their own boxes using the Lay out each of the graph’s groups in its own box option (Figure 4.6). Consider hiding edges between groups in one visualization to reveal and compare structure across groups, and creating a complementary overview visualization that shows connections across groups. These principles are based on the Netviz Nirvana guidelines developed by Shneiderman and colleagues [2, 3] and explored in the NodeXL environment by Bonsignore et al. [4].

ADVANCED TOPIC

Readable network layout heuristics Creating readable and accurate network graphs can be challenging. There is no single right way to lay out the vertices and edges in a network, just as there is no single way to photograph a landscape. However, several heuristics can be used to help assess the readability of your network layouts. • Make every vertex visible. If vertices overlap (i.e., occlude) each other, try reducing the scale of the network or spreading out the vertices (e.g., by modifying the strength of the repulsive force between vertices in the Fruchterman-Reingold layout as described in Section 4.4.3). • Make every vertex’s degree countable. Each edge emanating or ending at a vertex should be clearly visible. Spread out vertices, choose an appropriate scale, and make sure that edges don’t pass through vertices they are not connected to (i.e., avoid edge tunnels). When working with directed graphs (see Section 4.5), customize the size of the arrows so they don’t all blend together using the Graph Options>Edges>Arrow Size feature. • Make tracing edges from source to destination easy. Remove unnecessary edge crossings. If the network is small, consider using edge bundling, available through the Graph Options>Edges>Bundled drop-down menu. However, this can be very computationally intensive, so use with caution. When edge crossings are necessary, increase the angle at which edges cross so people don’t confuse two edges. Also avoid long edges when possible. • Make sub-groups and outliers identifiable. Separate out the network’s connected components (i.e., subnetworks that are not connected to one another by any

4.4.1  Manual layout You can manually position vertices to create arrangements that emphasize structures and create a more orderly display. You can even select multiple vertices by drawing a box around them or clicking on additional vertices while holding down the Control key. Multiple selected vertices move together when selected. Manually adjusting vertices is typically done to fine-tune existing networks that have been arranged automatically according to a layout algorithm.

4.4.2  Automatic layout NodeXL includes several automatic layout algorithms that can be applied to position all of the vertices according to a set procedure. The default layout algorithm for NodeXL is the Fruchterman-Reingold algorithm [5] described in the following section. In our example of the ABCD network, the Fruchterman-Reingold layout with its default settings is not particularly useful as it includes unnecessary edge crossings and a general sense of disarray that makes it difficult to identify important individuals (Figure  4.3). Use the drop-down menu directly above the graph pane or in the NodeXL Ribbon and select the Harel-Koren Fast Multiscale algorithm (Figure 4.5). This algorithm is a force-directed layout algorithm that attempts to minimize edge crossings [6]. Then click on the Lay Out Again button also above the graph pane. This will reposition all vertices into a structure similar to the one in Figure 4.4, though the specific orientation of that structure differs each time this algorithm is run.

II.  NodeXL tutorial: learning by doing

60

4.  Installation, orientation, and layout

FIGURE 4.5  Choosing the Harel-Koren Fast Multiscale layout algorithm from the drop-down menu. This menu also gives access to other layout algorithms including None and the Layout Options dialog shown in Figure 4.6.

Experimenting with different layout types (e.g., Circle, Sugiyama) can reveal useful patterns, relationships, or unusual features when you analyze a dataset.

4.4.3  Adjusting Fruchterman-Reingold settings The Fruchterman-Reingold Layout works well for many large social networks, though it may require some adjustment. It is an example of a force-directed algorithm, which uses an analogy of physical springs as edges that attract connected vertices toward each other and a competing repulsive force that pushes all vertices away from one another, whether they are connected or not [5, 7]. It typically results in edges that are relatively similar in length, though the length of edges has no specific meaning in most network visualizations. The algorithm uses an iterative process to adjust the placement of the vertices in order to minimize the “energy” of the system. Because it is an iterative layout, it runs many times, each time incrementally changing the position of each vertex based on the prior position. Adjust the Fruchterman-Reingold default parameters by selecting Layout Options… (Figure 4.5), which will open the Layout Options dialog (Figure 4.6). Set the Strength of the repulsive force between vertices to 1.5 and Iterations per layout to 20. Choose the Fruchterman-Reingold layout and click on Refresh Graph to see an updated version of the graph (Figure 4.7) that looks much more readable than the original, though perhaps still not as clear as the Harel Koren

FIGURE  4.6  Layout Options dialog showing the default settings for Fruchterman-Reingold layout.

FIGURE  4.7  ABCD network laid out with the FruchtermanReingold layout and adjusted parameters as described in Section 4.4.3.

II.  NodeXL tutorial: learning by doing



4.4  Layout: Arranging vertices in the graph pane

Fast Multiscale algorithm (Figure  4.4). Other Layout Options settings will be discussed in Chapter  7 since they relate to Groups.

4.4.4  Updating the graph pane Any time you change the underlying data or features that affect the layout of the network (e.g., directed vs undirected), you must click on the Refresh Graph button to update the graph. If you just want to change the layout, you can select a new layout type and click on Lay Out Again to reduce processing time. Note that the FruchtermanReingold layout will start from the current layout if you choose Lay Out Again, whereas it will start from an initial seed position if you choose Refresh Graph.

4.4.5  Preserving a layout Take a moment to manually position the vertices so they are in a more esthetically pleasing and readable graph (e.g., Figure 4.8). Once you have fine-tuned a graph layout like this, you can set the layout option to None in the automatic layout drop-down menu (Figure  4.5). Otherwise, if you choose to Refresh Graph, the vertices will be repositioned again according to the layout algorithm listed. The hard work of manually fine-tuning a layout can be lost since unfortunately there is no undo button for actions taken on the graph. A more permanent solution that will make sure others who open your file see your final layout requires a bit more work. First, reveal the Layout columns (see Advanced topic: Using hidden layout columns) on the Vertices worksheet. Next, change the value to Yes or 1 in the Locked? column for each

A DVANCED TOPIC

61

specify the X and Y coordinates of each Vertex. You can use formulas to set the values for these positions, thus creating your own layout algorithm. Or set the X properties of certain vertices so they line up perfectly along the X axis. As described in Section 4.4.5 you can use the Locked? column to fix the position of each vertex. This will override all automatic graph layouts so they don’t have any impact, so change it back if needed later. The Layout Order column specifies the order in which the vertex will be positioned on the graph. Use this column to make sure certain vertices are next to each other when using the Circle layout (e.g., sort the Layout Order column by some metric of interest such as Degree, which is described in Chapter 6).

­vertex (Figure  4.8). This will fix the X and Y coordinates of each vertex even if the file is shared with others or an automatic algorithm is run.

4.4.6  Graph pane tools You can customize network graphs using a variety of tools and options available via the NodeXL graph pane menu (Figure  4.9). The arrow on the left allows easy selection and deselection of multiple vertices. You can draw boxes around a set of vertices you’d like to move or manipulate. Right-clicking on a vertex will pull up many additional options, such as choosing Adjacent Vertices (i.e., the vertices connected to the one you are right-clicking). The plus and minus arrows zoom in and out from the clicked selection point. You can also zoom into the graph using the sliding bar, move around after zooming in by using the hand icon, and change the scale of the graph using the scale sliding bar.

4.4.7  Graph pane options

Using hidden layout columns NodeXL stores attributes of Edges, Vertices, and Groups in multiple columns of the corresponding worksheets. There are a large number of columns, so they have been grouped together by theme or function. For example, on the Vertices worksheet the Visual Properties columns store data related to the color, size, opacity, and shape of each Vertex (described in Chapter 5). Some columns, such as the Layout columns are hidden by default to reduce clutter. To hide or reveal column groups, select the Workbook Columns from the Show/Hide portion of the NodeXL ribbon and check (or uncheck) the categories you’d like to show (or hide) (Figure  4.8). Checking the Layout category reveals several layout related columns with data that can be manipulated. For example, columns

Choosing the Graph Options button on the graph pane toolbar will provide access to advanced features that allow you to change the default visual properties of edges and vertices (e.g., color, size, drop-shadows, glow), arrow size for directed graphs, and the curvature of edges (Figure  4.10). Change the default edge Width to 1.5 (Figure 4.10) to make the ABCD network graph more readable and esthetically pleasing. You can also add a background image (such as a map), as well as customize label fonts (including truncating text), including their position and color. When you make changes to Graph Options, they will affect all of your images. You can revert to the default settings using the Reset All button. NodeXL is a highly versatile and customizable network drawing program that can be adjusted according to individual needs.

II.  NodeXL tutorial: learning by doing

62

4.  Installation, orientation, and layout

FIGURE 4.8  Choosing to show Layout columns in the Workbook Columns menu on the NodeXL ribbon.

FIGURE 4.9  Graph pane toolbar.

4.5  Undirected and directed graph type So far, we have thought of the ABCD network as undirected. The edges were said to represent mutual friendship connections on LinkedIn. Thus, an edge either exists (for the pair of vertices it connects) or does not exist. Many edges are undirected, such as Facebook Friendships, affiliations (e.g., both people were tagged in a photo together), or mutual relationships (e.g., spouse). In contrast, other networks are directed, meaning that a vertex “points” to another vertex because of the nature of the link. Imagine that the ABCD network instead represents ABCD employees who have “endorsed” other ABCD employees on LinkedIn. The ABCD network would then be directed. Other directed networks are created when information is passed between people (e.g., a message is sent from one person to another), subscription relationships

(e.g., YouTube channel s­ ubscriptions; Twitter follower networks), invitation networks, etc. Notice that a single social media platform, such as LinkedIn, may have many different types of edges—thus many different networks, even among the same set of people (see Chapter 2). Visually, directed networks are displayed with arrows that point from the source vertex to the destination vertex. Additionally, certain network metrics (see Chapter  6) will be calculated differently for undirected and directed networks.

4.5.1  Changing the type of network To change the type of network, choose the appropriate Type of network from the drop-down list on the NodeXL ribbon menu in the Graph section. Change the type to Directed as shown in Figure 4.11 for the ABCD network. After you click on Refresh Graph you will

II.  NodeXL tutorial: learning by doing



63

4.6  Working with NodeXL files

sure to select the standard Excel Workbook (with a .xlsx extension). Do not save it as an Excel 97-2003 Workbook, a Macro-Enabled Workbook, or a Binary Workbook. All data, including the most recent layout positions will be saved. Save your ABCD network file now, since you will use it in later chapters.

4.6.2  Exporting an existing NodeXL file In addition, you can export NodeXL file data in several ways, accessible from the Export menu found in the Data section of the NodeXL Ribbon. Export options for common network file types such as UCINET, Pajek, GraphML, GEXF, and GDF are available. You can also export to an email, an automatically generated Power Point presentation, to the online network visualization tool called Polinode, or to the NodeXL Graph Gallery (Advanced topics: NodeXL graph gallery). Use the Export Options dialog to add metadata about your file.

A D VA NCE D TO P I C

NodeXL Graph Gallery FIGURE  4.10  Graph Options dialog showing the Edges options and Width set to 1.5.

see arrows that point from the people in Vertex1 to the people in Vertex2 (e.g., from Ben to Ava).

4.5.2  Reciprocated edges Directed edges can be reciprocated, meaning that two people may point toward each other. For example, if two Twitter users follow one another there would be two edges, each of which ends at a different user. In the ABCD network there are no reciprocated edges. If a directed edge was reciprocated, it would show up as a single edge between two vertices with an arrow at each end. In the spreadsheet there would be two rows, one for each directed edge. For example, currently there is an edge pointing from Ben to Ava so we see Ben in the Vertex1 column and Ava in the Vertex2 column in row 3 (Figure 4.11). If the edge were reciprocated, then another row would exist where Amy would be in the Vertex1 position and Bob would be in the Vertex2 position. Metrics for reciprocated edges are discussed in Chapter 6.

4.6  Working with NodeXL files 4.6.1  Saving NodeXL files To save the NodeXL file you have been working on, simply save it as you would any other Excel file making

NodeXL users are encouraged to use the NodeXL Graph Gallery found at https://nodexlgraphgallery.org to share datasets and visualizations with the community. You can export your files directly to the NodeXL Graph Gallery from the Export menu as described in Section 4.6.2. Use the Search bar (Figure 4.12) to find networks of interest and click on the image or link to see a close up version. Those uploading files can choose to share the underlying data (in GraphML format), NodeXL file, and NodeXL options used in the file. You can access them via links at the bottom of the detail page for the specific network.

4.6.3  Opening an existing NodeXL file You can open a NodeXL file in the same way you would normally open any Excel file. If NodeXL is installed, the NodeXL Ribbon and graph pane will automatically open when the file is opened.

4.6.4  Opening a NodeXL file created on another computer When opening a file that was created on an older version of NodeXL, you should open a blank NodeXL file and then use the Import>From NodeXL Workbook Created on Another Computer feature available via the NodeXL Ribbon.

II.  NodeXL tutorial: learning by doing

64

4.  Installation, orientation, and layout

FIGURE 4.11  The ABCD network after setting the Type to Directed, adjusting Graph Options as described in Section 4.4.7, Refreshing the graph, and manually moving vertices to create a more readable layout. Arrows point from vertices in the Vertex1 column toward vertices in the Vertex2 column.

4.6.5  Creating a trusted location for NodeXL files from the Internet Some Excel files can include malicious code. To prevent such files from causing trouble, recent versions of Windows require that you specifically give permission to open Excel (and thus NodeXL) files downloaded from the Internet. Permission can be given to a single file or to a folder, which can include as many files as desired. Instructions for creating a trusted location for various Window’s versions can be found on Microsoft’s support pages. For a single file you can right-click on the file, go to Properties, and choose Unblock if you have proper permissions as an administrator. To create a trusted folder try searching for “create, remove, or change a trusted location for your files” along with the name of your operating system. You are encouraged to create a NodeXL folder and

make it a trusted location where you store and access all your NodeXL files including those used throughout this book, which are available at https://www.smrfoundation. org/nodexl/teaching-with-nodexl/teaching-resources/.

4.7  Practitioner’s summary NodeXL is an Excel template that supports network analysis and visualization in a spreadsheet environment. This chapter introduced the basics of NodeXL using a simple corporate connection network. The spreadsheet includes an Edges worksheet that contains a row for each edge and a Vertices worksheet that contains a row for each vertex. Vertices can be arranged on the graph using a number of automatic layouts or can be manually positioned and fixed into place if desired. Files from a

II.  NodeXL tutorial: learning by doing



References

65

FIGURE 4.12  NodeXL Graph Gallery website for sharing NodeXL images and files.

variety of network tools can be imported and exported into NodeXL, which includes among the most customizable network visualization engines available.

4.8  Researcher’s agenda Designers of network-drawing programs must make many decisions as they develop their tools [8, 9]. Although traditional esthetic measures have emphasized making an appealing visual display and network layout [10–16], the contemporary view is more guided by user performance on benchmark tasks, such as comparing the degree of two vertices, identifying common properties of vertices, or spotting missing relationships [2, 17–19]. More sophisticated tasks such as finding cliques, cut vertices, or chains are also being used in studies. Increasing attention is given to avoiding overlap, making graphs readable, enabling users to follow edges from vertex to vertex, and showing connected components while pushing isolated vertices and small components to the side. The research questions become more challenging as the size and complexity of the network grow, leaving designers and users to make tradeoffs that may facilitate some tasks while making others more difficult [20]. Even more challenging is to have the visual design and network layout be done

automatically while producing an effective presentation [2, 21]. Fortunately, these topics are an active area of research, which is leading to clarification of what primary tasks users have for directed/undirected networks that emerge from differing application domains. User control is emerging as the solution, but that requires giving users a clear understanding of the implications of each choice. A growing body of literature on network visualization is found in information visualization journals and conferences and more specialized places such as the International Symposium on Graph drawing and Network Visualization.

References [1] R. Hanneman, M. Riddle, Chapter  5: Using matrices to represent social relations, in: Introduction to Social Network Methods, University of California, Riverside, Riverside, CA (published in digital form at http://faculty.ucr.edu/~hanneman). [2] B.  Shneiderman, A.  Aris, Network visualization with semantic substrates, IEEE Trans. Vis. Comput. Graph. 12 (5) (2006) 733–740. [3] C. Dunne, S.I. Ross, B. Shneiderman, M. Martino, Readability metric feedback for aiding node-link visualization designers, IBM J. Res. Dev. 59 (2/3) (2015) 14:1–14:16. [4] E.M.  Bonsignore, C.  Dunne, D.  Rotman, M.  Smith, T.  Capone, D.L. Hansen, et al., First steps to NetViz Nirvana: evaluating social network analysis with NodeXL, in: Proc. IEEE International Symposium on Social Intelligence and Networking (SIN-09) August, Vancouver BC, Canada, 2009.

II.  NodeXL tutorial: learning by doing

66

4.  Installation, orientation, and layout

[5] T.M.J.  Fruchterman, E.M.  Reingold, Graph drawing by force-­ directed placement, Softw. Pract. Ex. 21 (11) (1991) 1129–1164. [6] D.  Harel, Y.  Koren, A fast multi-scale method for drawing large graphs, J. Graph Algorithms Appl. 6 (3) (2002) 179–202. [7] M. Kaufmann, D. Wagner, Drawing Graphs: Methods and Models (Lecture Notes in Computer Science), Springer, 2001. [8] G.D.  Battista, P.  Eades, R.  Tamassia, I.G.  Tollis, Graph Drawing: Algorithms for the Visualization of Graphs, Prentice Hall, New York, 1998. [9] C.  Ware, Information Visualization: Perception for Design, third ed., Morgan Kaufmann, 2012. [10] M.  Smith, B.  Shneiderman, N.  Milic-Frayling, E.  MendesRodrigues, V. Barash, C. Dunne, et al., in: Analyzing (social m ­ edia) networks with NodeXL, Proc Communities & Technologies Conference, State College, PA, 2009. [11] D.  Archambault, H.C.  Purchase, Mental map preservation helps user orientation in dynamic graphs, in: International Symposium on Graph Drawing, Springer, Berlin, Heidelberg, 2012. [12] T. Dwyer, B. Lee, D. Fisher, K.I. Quinn, P. Isenberg, G. Robertson, et al., A comparison of user-generated and automatic graph layout, IEEE Trans. Vis. Comput. Graph. 15 (6) (2009) 961–968. [13] W.  Huang, S.-H.  Hong, P.  Eades, How people read sociograms: a questionnaire study, in: APVis ’06: Proc. 2006 Asia-Pacific Symposium on Information Visualisation, 2006, pp. 199–206. [14] W.  Huang, S.-H.  Hong, P.  Eades, Effects of sociogram drawing conventions and edge crossings in social network visualizations, J. Graph. Algorithms Appl. 11 (2) (2007) 397–429. [15] S. Kieffer, T. Dwyer, K. Marriott, M. Wybrow, HOLA: human-like orthogonal network layout, IEEE Trans. Vis. Comput. Graph. 22 (1) (2016) 349–358. [16] V.  Yoghourdjian, D.  Archambault, S.  Diehl, T.  Dwyer, K.  Klein, H.C. Purchase, H.Y. Wu, Exploring the limits of complexity: a survey of empirical studies on graph visualisation, Vis. Inform. 2 (4) (2018) 264–282.

[17] P. Eades, K. Sugiyama, How to draw a directed graph, J. Inf. Proc. 13 (4) (1990) 424–437. [18] W. Huang, Using eye tracking to investigate graph layout effects, in: APVis’07: Proc. 2007 Asia-Pacific Symposium on Information Visualisation, 2007, pp. 97–100. [19] F.  van Ham, B.E.  Rogowitz, Perceptual organization in ­user-generated graph layouts, IEEE Trans. Vis. Comput. Graph. 14 (6) (2008) 1333–1339. [20] C.  Ware, H.C.  Purchase, L.  Colpoys, M.  McGill, Cognitive measurements of graph aesthetics, Inf. Vis. 1 (2) (2002) 103–110. [21] T. Dwyer, Y. Koren, K. Marriott, IPSep-CoLa: an incremental procedure for separation constraint layout of graphs, IEEE Trans. Vis. Comput. Graph. 12 (5) (2006) 821–828.

NodeXL papers E.M.  Bonsignore, C.  Dunne, D.  Rotman, M.  Smith, T.  Capone, D.L.  Hansen, et  al., First steps to NetViz Nirvana: Evaluating social network analysis with NodeXL, in: Proc. IEEE International Symposium on Social Intelligence and Networking (SIN-09) August, Vancouver BC, Canada, 2009. D.L.  Hansen, D.  Rotman, E.M.  Bonsignore, N.  Milic-Frayling, E.M.  Rodrigues, M.  Smith, B.  Shneiderman, Do you know the way to SNA?: a process model for analyzing and visualizing social media data, in: Proc. ASE International Symposium on Social Informatics (SocialInformatics 2012), December 2012, Washington, DC, 2012. M.  Smith, B.  Shneiderman, N.  Milic-Frayling, E.  Mendes-Rodrigues, V. Barash, C. Dunne, et al., Analyzing (social media) networks with NodeXL, in: Proceedings Communities & Technologies Conference, State College, PA, 2009.

II.  NodeXL tutorial: learning by doing

C H A P T E R

5 Labeling and visual attributes O U T L I N E 5.1 Introduction

67

5.2 Labeling 5.2.1 Viewing attribute data in the ABCD network file 5.2.2 Labeling vertices 5.2.3 Adding tooltips 5.2.4 Formatting, positioning, and truncating labels using label options 5.2.5 Label vertex shape 5.2.6 Labeling edges

67

5.3 Visual properties 5.3.1 Vertex color 5.3.2 Vertex shape

71 72 73

Vertex size Vertex opacity Vertex visibility Edge visual properties Showing the graph legend Saving graph images and right-click graph menu 5.3.9 Graph options

68 68 69 70 70 71

5.1 Introduction Network data is often accompanied by data or textual information that describes each vertex and edge. Attribute data describing each vertex in a network is often available, particularly for social media datasets. Vertex data often includes information about usernames, profile information (location, biography, hometown), number of friends/followers, number of posts, account creation date, etc. Data about each edge can also include information about the type of edge (reply-to; mention; follow), content of a message (tweet; email content), number of messages, etc. Additional data about each edge and vertex can come from calculated network metrics (Chapter  6), group properties (Chapter  7), or textual content (Chapter  8) as described in later chapters. Numerous insights can be revealed by integrating vertex and edge metadata into network visualizations through the use of labels and visual attributes such as vertex size, color, and opacity or edge width. NodeXL provides a variety of options to add custom labels to vertices and edges, as well as change visual attributes of vertices and edges. Network analysts become artists and communicators as they create custom visualizations aimed at

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00005-4

5.3.3 5.3.4 5.3.5 5.3.6 5.3.7 5.3.8

74 75 75 75 76 76 76

5.4 Practitioner's summary

77

5.5 Researcher’s agenda

77

References

78

a­ ccurately and effectively representing underlying data, while also inspiring viewers.

5.2 Labeling Labeling vertices and edges is essential to creating readable graphs. Advanced topic: Labeling best practices shows some best practices for effectively using labels.

A D VA N C E D T O P I C S Labeling best practices

• Edge labels are very difficult to read and should only be used on small graphs with few edge crossings. Instead, use visual properties, such as color, width, and opacity of edges to differentiate between them. • Use vertex labels selectively for large graphs. Including a label for each vertex becomes problematic when there are over 50 or so vertices. Label only the most important vertices, or the ones that you are mentioning in related text.

67

© 2020 Elsevier Inc. All rights reserved.

68

5.  Labeling and visual attributes

• When labels are essential and the network is not too

•

•

•

•

large, consider making the label into the vertex by using the Shape>Label in visual properties (see Section 5.2.5). This creates a rectangular box around the label to which the edges connect, making the label’s connection to the vertex unmistakable (see Figure 5.6). Consider using images in place of labels for vertices where an image can more efficiently convey content (see Section 5.3.2). For example, many Twitter images have logos that convey a company more efficiently than a long company name. Additional information, such as demographic information or network metric values, can be added to labels. For example, label a vertex with “Dave (12)” to indicate a name and age. Truncate (i.e., shorten) labels when needed since long labels can hide edges and clutter the graph. Just make sure the labels are unique after truncation to avoid confusing one vertex for another. Use Tool Tips and custom right-click menus (see Advanced topic: Useful Excel formulas for labeling) to provide details on demand. For example, use a person's name for the label, but have their age show up when the cursor mouses over it. Or link to their user profile via a custom URL that appears when you rightclick on the vertex.

5.2.1  Viewing attribute data in the ABCD network file This chapter will rely on the ABCD network file you created in the last chapter (or download from https:// www.smrfoundation.org/nodexl/teaching-with-nodexl/teaching-resources/). Instructions for downloading the file and creating a Trusted Location for it in Windows were provided at the end of Chapter 4. This data file includes additional attribute data describing the edges and vertices, which will be mapped to labels and visual attributes. If you’d rather enter the data manually, the data values are available in Figures  5.1 and 5.2. Both the Edges (Figure 5.1) and Vertices (Figure 5.2) worksheets have an Other Columns section where a column of data can be added about edges or vertices. Data added immediately to the right of the furthest right column in the table will make the new data become part of the table as indicated by the column header turning blue. This is important, adding a column within the table makes the added data available to other parts of NodeXL that will be introduced later. Any type of data can be entered in these fields including textual, numerical, date/time, etc. Notice that the Years_at_ABCD column includes numerical data, while the other fields are textual.

FIGURE 5.1  ABCD network Edges worksheet showing additional data columns.

5.2.2  Labeling vertices Navigate to the Vertices worksheet of the ABCD network file. Scroll over to the Labels columns. Right-click on the first cell in the Label column, choose Format Cells…, and change it from Text to General. Then enter = [Vertex] into the cell and press enter so that it will reference the names that are found in the Vertex column (i.e., Column A) on the Vertices worksheet. Excel should copy this

FIGURE  5.2  ABCD network Vertices worksheet showing additional data columns.

II.  NodeXL tutorial: Learning by doing



5.2 Labeling

69

FIGURE 5.3  Sample ABCD network vertices with labels applied and a tooltip shown for Camila.

formula down the complete column so that all rows in the Label column have the same formula. Refresh Graph using the Karel-Koren layout to see the labels appear on the graph, as shown in Figure 5.3. By default labels show up underneath the vertex. However, this can be adjusted for each vertex using the Label Position column. Excel includes many textual formulas that can be used to help adjust labels. For example, Advanced topic: Useful Excel formulas for labeling explains formulas that can combine data fields into a single label or shorten long text fields. If you are not using formulas to populate the Label column, you could copy-and-paste values, type them in manually, or use Autofill Columns (see Advanced topic: Using autofill columns). The Autofill Columns feature is a core feature in NodeXL that will save considerable time and be used throughout most of the chapters in this book.

A D VA N C E D T O P I C

Useful Excel formulas for labeling One of the great benefits of NodeXL is that it can leverage all of the built-in features of Excel, such as formulas. Below are a few examples of useful formulas for creating meaningful labels. Make sure you reformat the column as General instead of Text before using these formulas. These formulas are based on the data in the ABCD network.

• Shorten names by using = LEFT([Vertex],3). This truncates (i.e., shortens) the names to 3 letters, or whatever number you place at the end of the formula, so that Camila becomes Cam. To visually indicate which names were truncated try using the following formula, which adds an ellipsis after only those that are truncated: =IF(LEN([Vertex])  > 3,LEFT([Vertex],3)&"…", [Vertex]). Camila now becomes Cam… while Ava remains Ava. • Combine data from different columns by using the & symbol along with quotes to indicate text. For example, =[Vertex]&" "&"("&[Role]&")" results in Ava (Manager), since she has Manager in the Role column. • Add a * to highlight specific names using = If. For example, to star all of the names with over 10 years working at ABCD use: =IF([Years_at_ ABCD] > 10,[Vertex]&"*",[Vertex]), which will result in Ava and Camila*. • Other useful formulas to look into include Clean, Trim, Upper, Proper, Lower, Len, Iferror, and the use of "" and & in text strings (as in the prior bullet).

5.2.3  Adding tooltips To reduce information clutter, some information can be displayed only when the mouse is placed over a specific vertex. This is called a tooltip. For example, in Figure 5.3, the

II.  NodeXL tutorial: Learning by doing

70

5.  Labeling and visual attributes

cursor was placed over Camila’s vertex, which prompted her role (Manager) to appear nearby. To add a tooltip, populate the Tooltip column on the Vertices worksheet. This can be done using copy-and-paste, manually, via formulas, or using the Autofill Columns feature as described in Advanced topic: Using autofill columns.

A D VA N C E D T O P I C Using autofill columns

The Autofill Columns feature of NodeXL allows you to automatically populate (or delete) information in the Edges, Vertices, and Groups worksheets. To use the feature, click on the Autofill Columns icon in the Visual Properties section of the NodeXL Ribbon, which opens a new dialog as shown in Figure 5.4. Notice there are three tabs at the top, each of which corresponds to a different worksheet in the NodeXL workbook. On the left of this dialog is a column of all the visual properties, labeling, and positioning attributes. In the second column, choose the field with the data that you want to use to govern this attribute from the drop-down menu. For example, in Figure 5.4 choose Role and click Autofill. This will populate the Vertex Tooltip column of the Vertices worksheet with the data in the Role field and refresh the graph. The Options arrow next to each field allows you to clear the associated worksheet column, or change other properties related to the mapping that are discussed later in this chapter.

5.2.4  Formatting, positioning, and truncating labels using label options Several additional visual attributes of labels can be modified through the Label Options dialog. Click on the Graph Options button on the NodeXL Graph Pane (highlighted in Figure  5.5) to open the Graph Options dialog. Navigate to the Other tab, and click on the Labels… button also highlighted in Figure 5.5. This will open the Label Options dialog shown at the bottom of Figure 5.5 where you can automatically truncate labels, change the font and textual properties of labels, set the default position of labels in comparison to the vertices, and more. These can be adjusted for edge, vertex, or group box labels (discussed in Chapter 7).

A D VA N C E D T O P I C Visual property best practices

Not all visual properties are created equal. The human eye perceives some more easily than others. Understanding when to use a specific property or when to combine properties is a mix of science and art. Below are some best practices to keep in mind:

• Avoid using too many visual properties at once,

• • • • • • •

particularly if they all map to different data. Instead, create multiple graphs of the same data and place them side-by-side (with the same layout) for easy comparison. Combine shape and color to emphasize distinct categories of vertices. For example, use orange disks for teachers and blue solid squares for students. Combine edge width and opacity on weighted edges, making sure the minimum and maximum widths and opacity ranges are set so they are all visible. Avoid combining color and opacity, since colors will look different when their opacity changes. Use size (or width) for numerical data, since humans can more accurately compare different sizes (or widths) than different gradations of color or opacity. Use solid shapes (e.g., disk) by default. Use outline shapes (e.g., circle) only if there are many vertices that are occluded (i.e., hidden by other vertices). Avoid using color if the image is likely to be printed in black and white. Use colors that those who are colorblind can differentiate.

5.2.5  Label vertex shape

FIGURE  5.4  Autofill Columns dialog showing the Vertices tab. The Vertex Tooltip is mapped to the Role data column.

To increase readability, it is often useful to turn the vertex into a label rather than the default disk (i.e., filled in circle). To do this, navigate to the Shape column on the Vertices worksheet. Place the cursor inside of cell C3 and a drop-down menu option will appear next to the cell on the right. Select the drop-down menu and scroll

II.  NodeXL tutorial: Learning by doing



71

5.3  Visual properties

FIGURE 5.5  Label Options dialog (bottom) and pathway to open it (buttons highlighted in yellow).

down to choose Label (see Figure 5.6). Copy this down for all vertices and click Refresh Graph, which will show an updated graph like the one shown in Figure 5.6. When the Shape is set to Label, other visual properties such as color and size still apply to the vertex (see Section 5.2). The background color of the box surrounding the label can be set differently for each Vertex using the Label Fill Color column.

However, when data is qualitative or unique, and the network size is small, edge labels can be useful. Adding label text to the Label column on the Edges worksheet is similar to adding it to the Vertices worksheet. You can also customize the color and size of the edge label by entering data into the Label Text Color and Label Font Size columns on the Edges worksheet.

5.3  Visual properties

5.2.6  Labeling edges Edges can also be labeled, though this is less common than labeling vertices, because edge labels are difficult to read on most networks. Typically, other visual properties, such as width or color, can represent the value or type of an edge more effectively than edge labels.

NodeXL is a sophisticated and flexible network visualization tool, allowing you to map many types of data to a variety of visual properties of a network graph. For example, the color of a vertex may be based on demographic data such as gender or age. Or the size of a vertex may be

II.  NodeXL tutorial: Learning by doing

72

5.  Labeling and visual attributes

FIGURE 5.6  Results of setting the Vertex Shape to the Label option.

based on a network metric such as Degree or Betweenness Centrality (see Chapter 6). A combination of different visual attributes can be used to help draw attention to different details. For best practices related to visual properties see Advanced topic: Visual property best practices. The Vertices worksheet includes a set of columns grouped under Visual Attributes including Color, Shape, Size, Opacity, Image File, and Visibility. Figure 5.7 shows the many visual attributes that can be applied to vertices. Values for each visual attribute can be typed into the spreadsheet manually, populated via a formula, selected from a drop-down that shows up when the cursor is inside of a cell (e.g., the Shape column), selected from the Visual Properties menu ribbon items, or automatically filled in based on the Autofill Columns feature (see Advanced topic: Using autofill columns). Some effects, such as Glow, Drop Shadow, and Selected color are determined in the Graph Options dialog (see Advanced topic: Graph options).

blue and the female students to a custom color. To set Ben’s color, type Blue into the Color column on Ben’s row. You can type in any color from the 140 Cascading Style Sheet color names (a Google search will list them for you). Alternatively, you can choose a color from the

5.3.1  Vertex color To make the ABCD network graph more visually meaningful, change the color of the male students to

FIGURE 5.7  Vertex visual property examples.

II.  NodeXL tutorial: Learning by doing



5.3  Visual properties

color picker available in the NodeXL menu ribbon under the Visual Properties section (see highlighted menu button in Figure  5.8). If you choose Define Custom Colors and pick your own color, the spreadsheet will show the color’s 3-digit red-green-blue (rgb) number such as 230, 101, 6 in the spreadsheet cell (see Figure 5.8). Choose Refresh Graph to see the changes. Rather than manually entering colors, you could write an = IF() formula that sets the color in the Color column based on data in the Gender column. This is much faster than manually entering the data, particularly as datasets grow beyond a few dozen edges. Enter the following formula = IF([Gender] = "Male," "Blue," "230, 101, 6") and copy it to each of the cells in the Color column. Click Refresh Graph to see the changes take effect. An alternative method to set the vertex color is the Autofill Columns feature (see Advanced topic: Using autofill columns). The Vertex Color Options dialog lets you choose between two types of data: Categories or Numbers. Categorical data has distinct categories, such as the Gender column that includes the categories

73

of Male and Female. If you choose this option you cannot choose the specific colors that are chosen by NodeXL for each category, so using a formula does give you more control than this approach. Alternatively, numerical data can be used. If chosen, the raw numerical data (e.g., the Years_at_ABCD column) maps to a variety of colors that blend two colors selected by the user in the Vertex Color Options dialog.

5.3.2  Vertex shape The Vertex Shape column was first introduced in Section 5.2.5, when we set the Shape of each Vertex to Label. A variety of additional vertex shapes are available: solid shapes (Disk, Solid Square, Solid Diamond, and Solid Triangle), outline shapes (Circle, Square, Diamond, Triangle), and others (Sphere, Label, and Image). The Image shape only works if the Image File field is populated with a valid path name to a file on your computer (e.g., C:\ MyImages\Image.jpg) or a URL (e.g., http://www. somesite.com/Image.jpg). Some NodeXL network data

FIGURE 5.8  Changing vertex color using the color picker accessible via the menu ribbon.

II.  NodeXL tutorial: Learning by doing

74

5.  Labeling and visual attributes

importers, such as the Twitter importers, download user images and automatically populate the Image File field so that profile images can be used to represent each vertex. If the URL’s become broken links at a later time, a default image with a red X will be shown. If you have different types of vertices (e.g., students and faculty; wiki pages and wiki editors), you may want to use shape to differentiate between them. This can be done using formulas for categorical data. For numerical data, the Autofill Columns feature can be used to identify shapes automatically based on specific values (e.g., data that is greater than 10 will be a Solid Square, otherwise it will be a Disk).

5.3.3  Vertex size Similar approaches can be used to fill in the data for the Vertex Size column. When working with numerical data, such as the data in the Years_at_ABCD column, it is often useful to use the Autofill Columns feature of NodeXL to map the raw data onto the visual properties (e.g., Size). Open the Autofill Columns dialog, choose Years_at_ABCD from the drop-down menu next to Vertex Size, and then open the Vertex Size Options dialog as shown in Figure 5.9. The options dialog allows you to change details about the mapping of

the raw data onto the visual property data. For example, as shown in Figure 5.9, the Vertex Size Options dialog allows you to change the minimum and maximum size of the vertex. Change the maximum vertex size to 50 to increase the difference in sizes between the vertices. By default, a linear mapping is used. For example, Fay has the most years at ABCD (29) and Liu and Matt have the fewest (1). Notice that in the Size column, Fay has the maximum size (50) and Matt has the minimum size (1.5). All other employees are assigned Size values between these extreme values based on a linear mapping. This works well for this network, but for other networks you may want to choose the Ignore Outliers and/or Use a logarithmic mapping options on the Vertex Size Options dialog (see Figure 5.9). Outliers are identified as values that are at least one standard deviation above or below the average value of the raw data. Ignoring them will still include the vertex in the graph, but will not include the vertex’s value when calculating the value of the other vertices. Using a logarithmic mapping is useful when the raw data follows a logarithmic or power-law distribution, which is common in social media participation data (e.g., number of followers or posts). More advanced mappings can be performed using Excel formulas that populate the vertex property field (e.g., Size) based on the raw data field (e.g., Years_at_ABCD).

FIGURE 5.9  Using Autofill Columns to set the vertex size based on Year_at_ABCD data.

II.  NodeXL tutorial: Learning by doing



75

5.3  Visual properties

5.3.4  Vertex opacity Vertex Opacity determines the level of transparency (i.e., how see-through) for each vertex. Values can be between 0 (fully transparent) and 100 (fully opaque). The default value is 100. The Autofill Columns options allow you to determine the minimum and maximum value, similar to the Vertex Size Options dialog shown in Figure 5.9.

5.3.5  Vertex visibility When working with large networks, it is often useful to filter out some vertices, so they do not show up in the network. The Visibility column allows you to do so without deleting the information from the Excel spreadsheet. There are four options available. Show if in an Edge will display the vertex on the graph if the vertex is connected to another vertex by at least one edge. Otherwise, the vertex row will be ignored. This is the default. Skip will ignore the vertex row and any edges connected to it. It is as if the data is not in the spreadsheet, so graph metrics (see Chapter  6), groups (see Chapter 7), and the graph itself will not use the data present in any “skip” row. Hide will include the vertex in calculations for graph metrics, groups, and even use it to determine the positioning of other vertices in the graph, but will not display it. This is equivalent to setting its opacity (and the opacity of any edges associated with it) to 0. Show will assure that the vertex is always included, even if it has no edges connected to it.

5.3.6  Edge visual properties The Visual Properties columns on the Edges worksheet are slightly different, but work in a similar manner to the Visual Property columns on the Vertices worksheet. Figure 5.10 presents the many edge visual properties available in NodeXL. Color and Opacity work the same way as the corresponding vertex attributes. Style changes the type of line (Solid, Dash, Dot, Dash Dot, and Dash Dot Dot) and is comparable to the Shape column for vertices. It is best used when working with categorical data. Visually, different styles are difficult to differentiate in large networks, so coupling style with distinct colors is often useful. Width determines how wide the edge is and is most comparable to the Size vertex property. The Visibility column affects the visibility of edges and can be set to Show (always show, no matter what), Skip (act as if the edge does not even exist in the dataset), or Hide (do not display on the graph, but otherwise treat it as if it is present). See Chapter  7 for more examples of using the Visibility column to filter out edges or vertices. Additionally, the Graph Options allow you to create

FIGURE 5.10  Edge visual property examples.

Curved edges and Bundled edges (see Advanced topic: Graph options). Combining Size and Opacity when using numerical data can make differences between edges more distinct. Use the Autofill Columns feature to set the edge Width and Opacity based on the Shared_ Connections column as shown in Figure  5.11. This represents the number of shared friends that each pair of people have. Change the minimum edge opacity to 50 as shown in Figure 5.11. Also change the Edge Width Options to have a minimum of 1.5 and a maximum of 5. This will assure that each edge is visible, but not too wide. After clicking Autofill, the graph should look similar to the one shown in Figure 5.11.

A D VA N C E D T O P I C

Exporting and importing NodeXL options It can take considerable time to customize graphs so they look the way you desire. NodeXL allows you to import and export options settings (e.g., visual properties, labels, default settings), so you can use the same ones in different workbooks. To do so, find the Options portion of the NodeXL Ribbon. Clicking on Export will let you save down a .NodeXLOptions file that you can name with the options settings for the current workbook. Choosing Import will allow you to import such options into a workbook. You can also Use Current for New, which will use the current workbook's options for all new NodeXL workbooks. Reset All will reset all options to the original defaults. NodeXL has “recipes.”

II.  NodeXL tutorial: Learning by doing

76

5.  Labeling and visual attributes

FIGURE 5.11  Using Autofill Columns to set the Edge Width and Edge Opacity based on Shared_Connections data on the Edges worksheet.

5.3.7  Showing the graph legend A graph legend can be included at the bottom of the image, as is done in Figure  5.12. To view the legend, check the Legend item in the Graph Elements dropdown menu found in the Show/Hide section of the NodeXL Ribbon (see Figure 5.12). Notice that color is not shown in the legend. This is because a formula was used instead of the Autofill Columns feature.

5.3.8  Saving graph images and right-click graph menu Right-clicking on the graph pane, or a specific vertex in the graph pane, will open up a customized menu as shown on the left-hand side of Figure 5.12. Menu items allow you to select and deselect subsets of vertices (e.g., adjacent vertices, or those that are connected to the selected vertex), edit the visual properties of selected edges or vertices, modify the layout, and adjust the layout. To save a graph (and legend if you have one showing), choose the Save Image to File option in the menu (as shown in Figure  5.12). You can modify the Image Options, which allows you to change the size of the graph pane in the created image, as well as add or remove a custom header and footer. When you choose

Save Image…, you will be prompted to choose a location and image file type. If you plan on printing the image, you may want to export it as an XPS file, which is a vector file format that can be scaled up or down to any size. The other file types are all pixel-based and will not scale well but may be well suited to web and small print contexts.

5.3.9  Graph options Options used in a current file can be shared across workbooks as described in the Advanced Topic: Exporting and importing NodeXL options. NodeXL allows you to customize many aspects of the graph pane through the use of the Graph Options dialog available on the menu at the top of the graph pane. There are three tabs in the dialog, each of which are described below. • Edges tab. Default edge color, size, arrow size, and opacity can be set. Additionally, edges can be made so they curve or are “bundled” (i.e., clustered together when many point to the same node), though bundling edges can slow down graph layouts considerably. • Vertices tab. Default color, shape, size, size of images, and drop shadow and glow effects can

II.  NodeXL tutorial: Learning by doing



77

5.5  Researcher’s agenda

FIGURE 5.12  Final ABCD network graph with graph legend shown.

be set (see Figure 5.7). Glow effects slow down graph layouts, but can look nice when using a dark background color (see next bullet). • Other tab. Background color or a background image can be set. For example, an image of a geographical map can be set as the background and nodes can be overlaid on top. Labels can be customized as described in Section 5.2.4. Additionally, custom right-click menu items can be added. For example, a url can be provided and a tip to open the URL when selected in the menu.

5.4  Practitioner's summary Network visualizations come to life when they are combined with other data describing the vertices and edges. NodeXL supports vertex and edge labels. Three

types of vertex labels can be used including: (1) adjacent labels that appear next to vertices on the graph, (2)  shape labels that replace the vertex with the label, and (3) tooltip labels that only appear when mousing over a vertex. Label position, color, font type, etc., can be customized using label options. Many visual properties can be mapped onto vertices including color, shape (including labels and images), size, and opacity. Visual properties for edges include color, width, style, and opacity. Features such as Autofill Columns and built-in Excel formulas can be used to enter data into the Visual Properties fields automatically.

5.5  Researcher’s agenda Creating network visualizations that help people gain insights from networks, particularly large and complex

II.  NodeXL tutorial: Learning by doing

78

5.  Labeling and visual attributes

networks, is an active area of research. There is a long history of research on information v ­isualization that identifies the visual properties (e.g., color, distance, size) that humans are most (or least) adept at understanding [1]. Most network visualization tools now allow attribute data to be mapped onto visualized attributes such as size, color, and shape. The combination of network data with attribute data is typically called multivariate network visualization [2], an active area of research given the difficult problems associated with such rich datasets. Researchers are increasingly examining richer visualizations for nodes including images, pie charts, or content-specific visuals such as 3D proteins [3]. Network visualization tools have also begun to integrate traditional node-link visualizations with alternative, complementary visualizations. For example, CyToStruct integrates node-link diagrams with three-dimensional molecular views important for bioinformatcs data [4], and NodeTrix integrates node-link diagrams with adjacency matrices that highlight local communities with social networks [5]. Other ­ content-specific network

­isualizations utilize rich sets of visual attributes or v symbols to help represent ­attribute data, such as in the Interactive Tree of Life (iTOL) viewer [6].

References [1] W.S.  Cleveland, R.  McGill, Graphical perception and graphical methods for analyzing scientific data, Science 229 (4716) (1985) 828–833. [2] A. Kerren, H.C. Purchase, M.O. Ward, Multivariate network visualization, in: Lecture Notes in Computer Science, Dagstuhl Seminar #13201, Springer, 2013. [3] M.E. Smott, K. Ono, J. Ruscheinski, P. Wang, T. Ideker, Cytoscape 2.8: new features for data integration and network visualization, Bioinformatics 27 (3) (2011) 431–432. [4] S. Nepomnyachiy, N. Ben-Tal, R. Kolodny, CyToStruct: augmenting the network visualization of cytoscape with the power of molecular viewers, Structure 23 (5) (2015) 941–948. [5] N.  Henry, J.D.  Fekete, M.J.  McGuffin, NodeTrix: a hybrid visualization of social networks, IEEE Trans. Vis. Comput. Graph. 13 (6) (2007) 1302–1309. [6] I. Letunic, P. Bork, Interactive tree of life (iTOL) v3: an online tool for the display and annotation of phylogenetic and other trees, Nucl. Acids Res. 44 (W1) (2016) W242–W245.

II.  NodeXL tutorial: Learning by doing

C H A P T E R

6 Calculating and visualizing network metrics O U T L I N E 6.1 Introduction

79

6.2 ABCD network example

79

6.3 Computing graph metrics 6.3.1 Overall graph metrics 6.3.2 Vertex-specific metrics

80 80 83

6.4 Marvel cinematic universe network example 84 6.4.1 Visualizing and interpreting metrics in a bimodal network 84 6.4.2 Mapping graph metrics to X and Y coordinates 88

6.1 Introduction When trying to understand networks, analysts often want to identify important vertices, locate subgroups, or get a sense of how interconnected a network is compared to other networks. Although visualization itself can help do this, it is often helpful to use the rich set of quantitative network metrics, also called network graph metrics, which have been developed by social network analysis researchers (Chapter 3). Network graph metrics can describe an entire network, subgroups, or specific actors within a single network. Aggregate graph metrics such as network density can be used to systematically compare communities, helping analysts decide which communities are highly connected and which are sparse. Tracking aggregate graph metrics over time can determine the effectiveness of interventions on the network as a whole. For example, you would expect the total number of edges to grow, increasing the density of the graph, after a photo sharing activity designed to introduce people to those they don’t know. Individual person-level metrics provide insights about a person’s position within the network, helping to identify important or “central” people. For example, network graph metrics help identify people who are bridge spanners or who are popular in a network. Once identified, analysts and managers can better know who to contact or

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00006-6

6.5 CSCW 2018 conference Twitter network example 6.5.1 Calculating and interpreting directed network metrics 6.5.2 Examining top items output 6.5.3 Examining time series output

88 89 91 92

6.6 Practitioner’s summary

93

6.7 Researcher’s agenda

93

References

93

influence or bring to the table when trying to implement new programs or gain broader understanding. Metrics can also be used to identify cliques or persistent social roles that show up in many communities. Understanding the mix of social roles that exist within a particular network can help analysts determine if they have a healthy mix of social types or who may be a good candidate to replace an outgoing leader. NodeXL calculates several network graph metrics. Once calculated, you can use these metrics to change the visual display of your network graphs in powerful ways as shown in this chapter. You can also filter out vertices or edges based on network metrics as discussed later (Chapter 7).

6.2  ABCD network example To better understand the meaning of each graph metric, start by opening the ABCD network visualized in the last chapter (or download it from https://www.smrfoundation.org/nodexl/teaching-with-nodexl/teaching-resources/). This network was designed specifically to illustrate the differences between several key metrics. If you download the version online, the layout shown in the book will be reproduced since the vertex positions are locked in place - the Locked? column values are set to

79

© 2020 Elsevier Inc. All rights reserved.

80

6.  Calculating and visualizing network metrics

Yes (see Advanced topic: Using hidden layout columns in Chapter 4). It should also be set to an undirected network type. Instructions on importing a NodeXL file made on another device can be found at the end of Chapter 4.

6.3.1  Overall graph metrics Navigate to the Overall Metrics worksheet, which summarizes some of the key properties of the entire network (Figure 6.2). These metrics include the following:

To calculate graph metrics, first click on the Graph Metrics button on the Analysis section of the NodeXL ribbon. This opens the Graph Metrics dialog (Figure 6.1). Select the metrics you want to calculate by checking in the boxes next to them. Details about the metric that is selected (e.g., Vertex clustering coefficient) are shown in the box below (see Figure 6.1). Some metrics allow you to customize various options by clicking on the Options… button on the right-hand side. Check the boxes next to the metrics shown in Figure 6.1 and then click Calculate Metrics. Some of the graph metrics can take a while to calculate when working with large networks, so a status bar is used to show progress. NodeXL will create a new Overall Metrics worksheet and take you there to show summary information for the entire network. It also populates a set of Graph Metrics columns on the Vertices worksheet that shows vertexspecific metrics, such as centrality metrics.

• Graph type. Undirected or directed. • Vertices. The number of total vertices (i.e., rows on the Vertices worksheet). • Unique edges. The number of unique edges found in the Edges worksheet. • Edges with duplicates. The number of repeated vertex pairs on the Edges worksheet. Duplicate vertex pairs may occur, as, for example in a Twitter network when person A mentions person B in multiple tweets. Duplicate vertex pairs are treated as a single edge for most metrics, since NodeXL metrics currently do not support weighted networks. The ABCD network does not include any duplicate edges. • Total edges. The number of total edges (i.e., rows on the Edges worksheet), which is the sum of the Unique Edges and Edges With Duplicates. • Self-loops. The number of edges that connect a vertex with itself. A self-loop occurs when the edge list includes the same exact name in the vertex 1 and

FIGURE 6.1  The Graph Metrics dialog with checks next to relevant

FIGURE  6.2  Part of the Overall Metrics worksheet showing data

metrics.

for the ABCD network.

6.3  Computing graph metrics

II. NodeXL tutorial: Learning by doing



6.3  Computing graph metrics

vertex 2 columns on the Edges tab (i.e., a person is connected to themselves). This may happen when, for example, in an email list a person replies to his or her own email. Self-loops are represented visually in the graph pane by a circular edge that comes out of a vertex and returns to that same vertex. Metrics do not include self-loops in their calculations unless otherwise stated in the description of the metric (e.g., see degree, in-degree, and out-degree). • Reciprocated vertex pair ratio. This is only applicable for directed networks. It represents the percent of connected vertex pairs that are connected by directed edges pointing in both directions. It is calculated as: (the number of vertex pairs that are connected to each other in both directions)/(the number of vertex pairs that are connected by one or two edges). In a Twitter network, where edges represent Mentions, a high Vertex Pair Ratio indicates that most of the time when a user mentions a second user, the second user also mentions the first user. • Reciprocated edge ratio. This is also only applicable for directed networks. It represents the percentage of edges that are reciprocated (i.e., an edge that has a companion edge pointing in the opposite direction between the same two vertices). It is calculated as: (the number of edges that are reciprocated)/(the number of total edges). While this value is correlated with the Reciprocated

81

Vertex Pair Ratio, it is not the same. It will typically be higher because a single reciprocated vertex pair consists of two reciprocated edges, making the numerator twice as high for the reciprocated edge ratio. While the denominator is also higher for the reciprocated edges calculation, it won’t be double that of the reciprocated vertex pair ratio. • Connected components. The number of connected components (i.e., clusters of vertices that are connected to each other but separate from other vertices in the graph). In the ABCD network there is only one connected component because you can get from one vertex to all other vertices. If Ava was not connected to Ethan, then there would be two connected components (see Figure 6.3). • Single-vertex connected components. The number of isolated vertices that are not connected to any other vertices in the graph. There are no isolated vertices in the ABCD network. If Dmitri was not connected to Ava, he would not be connected to anyone in the network and would then become a single-vertex connected component. • Maximum vertices in a connected component. The number of vertices in the connected component with the most vertices. This is equal to the number of vertices (13) in the ABCD network, because they are all part of the only connected component.

FIGURE 6.3  The ABCD network showing graph metrics for each vertex. Degree is mapped to size (2–100), betweenness centrality is mapped to opacity (65–100), eigenvector centrality is a tooltip, and Shared_Connections are mapped to edge weight (1.5–5) and edge opacity (50–100).

II. NodeXL tutorial: Learning by doing

82

6.  Calculating and visualizing network metrics

• Maximum edges in a connected component. The number of edges in the connected component with the most edges. This is equal to the number of edges (18) in the ABCD network, because they are all part of the only connected component. • Maximum geodesic distance (diameter). The geodesic distance is the length of the shortest path between two people. If you think of the edges as roads and the vertices as houses, the geodesic distance would be the number of roads someone must take to get from one house to another, assuming that the person is traveling on the shortest path possible. The maximum geodesic distance, or diameter of a network, is the largest geodesic distance in the network, or the distance between the two vertices that are farthest from each other. In the ABCD network, this value is 4. For example, the shortest path between Liu and Kate is 4; similarly the shortest path from Camila to Ji-yoo, Hassan, Matt, and Kate is also 4. All other geodesic distances are smaller. For example, the shortest path between Gabe and Fay is 1. • Average geodesic distance. This is the average length of the geodesic distances between all pairs of vertices. It gives a sense of how “close” community members are from one another. For example, in the ABCD network, a value of 2.22, suggesting that many people in the network know others directly or through a friend of a friend. Interestingly, many large social networks retain a relatively small average geodesic distance. For example, the Facebook network showed an average geodesic distance of only 4.57 with over 1.59 billion Facebook users in 2016.1 • Graph density. The graph density is a number between 0 and 1 that indicates the percentage of possible edges that are realized. It is a measure of how interconnected the vertices are in the network. The specific formula for calculating graph density is: (number of actual edges in the network)/(number of possible edges in the network). For undirected graphs the numerator is multiplied by 2. The number of possible edges in the network is based on the total number of vertices (n) in the network. Specifically, it is n*(n − 1). For example, the numerator for the ABCD network is 36 (i.e., 2*18) since it is an undirected network with 18 edges. The denominator is 156 (i.e., 13*12) since there are 13 vertices. Thus, the Graph Density is 0.23 (36/156). If more of the 18 employees became connected, this number would increase. Larger social networks tend to have lower graph densities, all things being equal, so be careful comparing this metric across different networks.

• Modularity. The modularity metric is only calculated when working with subgroups, which are discussed in Chapter 7. Modularity measures how distinct different subgroups of vertices are from the rest of the network [1]. It can be used to measure the “quality” of the separation of vertices into subgroups. Networks with high modularity have dense connections between vertices that are part of the same subgroup (i.e., module), but sparse connections between vertices that are part of different subgroups. More specifically, modularity is the fraction of within-group edges minus the expected fraction of edges if edges were distributed at random. It will be a value between −1 and 1, where positive values indicate that the number of within-group edges is higher than the expected number of edges based on chance. • NodeXL version. Indicates the version of NodeXL in use when metrics were calculated. In addition, a frequency chart is created for each of the possible vertex-specific graph metrics. These frequency charts are particularly helpful when analyzing large networks. Some basic statistics about the metric distributions are shown under the charts (minimum degree, maximum degree, average degree, and median degree). These help characterize the entire networks and allow for comparisons over time or across networks.

A D VA N C E D T O P I C

Calculating and importing additional graph metrics Numerous network metrics exist in addition to those calculated by NodeXL. Furthermore, new metrics are constantly being developed. Additional aggregate metrics can be calculated using Excel’s built-in functions. For example, some analysts like to look at the variance of the degree, which can be calculated by using the function: =VAR.P(Vertices[Degree]) Other aggregate metrics capture aspects of the degree distribution. For example, network centralization measures how much the network depends on key people for its connectivity. Many of these aggregate metrics can be calculated in NodeXL using formulas. Some specialized graph metrics are not currently calculated in NodeXL, such as centrality metrics that use edge weights (see Newman’s thoughtful review [2] for a comprehensive discussion of network metrics). Researchers who need such metrics may use other network analysis tools such as Pajek or UCINET to calculate them and import them into NodeXL as additional columns. This allows all of the advanced visualization features of NodeXL while still providing more network metrics.

1 https://research.fb.com/three-and-a-half-degrees-of-separation/.

II. NodeXL tutorial: Learning by doing



6.3  Computing graph metrics

6.3.2  Vertex-specific metrics The different vertex-specific metrics, also called centrality metrics, help identify who is “important” or “central” to a network. Of course, people are important in different ways. Some may have the most direct connections, while others may be important bridge spanners who connect otherwise disparate parts of the network. Each centrality metric captures a different aspect of importance as described below. To see the vertex-specific metrics navigate to the Vertices worksheet. You will see the new Graph Metrics columns, which can be hidden later if desired by unchecking Graph Metrics from the Workbook Columns button on the NodeXL ribbon. Each value relates directly to one of the vertices. For example, row 4 shows the graph metrics that are specific to Ava (Figure 6.3). Vertex metrics can be mapped onto visual attributes (Figure 6.3), which you can recreate by using the Autofill Columns feature found in the NodeXL Visual Properties menu ribbon. The graph legend shows that degree (1–6) is mapped to size and betweenness centrality is mapped to opacity. Edge weight and opacity are also mapped to Shared_Connections (see Chapter 5). In addition, eigenvector centrality is mapped to the tooltip (see Ethan’s score in Figure 6.3) and the labels are set to Vertex and positioned so they don’t cross edges. A description of each metric and how it relates to the ABCD Network are provided below. It is often useful to sort the spreadsheet columns based on graph metrics. For example, the rows in Figure  6.3 are sorted based upon data in the Degree column as indicated by the downward pointing arrow inside the Degree drop-down menu. Degree The degree of a vertex (sometimes called degree centrality) is a count of the number of unique edges that are connected to it. Fay has a degree of 6 because she is directly connected to 6 other individuals. In comparison, Kate has a degree of only 1 because she is connected to only one other person. If the edges represented strong friendship connections between employees at ABCD, we might say that Fay is the most popular person in the network and Kate is one of the least popular. If you were analyzing a directed graph, the single degree metric would be split into two metrics: (1) In-degree, which measures the number of edges that point toward the vertex of interest (i.e., number of people who have received endorsements from others), and (2) Out-degree, which measures the number of edges that the vertex of interest points toward (i.e., the number of people the person has endorsed). In the ABCD network, NodeXL only calculates degree since the network is specified as containing undirected ties.

83

Betweenness centrality Although popularity is important, it is not everything. Betweenness centrality is a measure that captures a completely different type of importance: the extent to which a certain vertex lies on the shortest paths between other vertices. In other words, it helps identify individuals who play a “bridge spanning” role in a network. Consider Ethan in the ABCD network. He is directly related to only four people (i.e., he has a degree of 4). Despite his relatively low degree, his position as a “bridge” between Ava (and indirectly all those who Ava is connected to) and the rest of the group may be of utmost importance. If, for example, information were passed from one person to another, Ethan and Ava would be vital for assuring that Dmitri, Liu, Camila, and Ben could communicate with the rest of the group. In fact, if either Ethan or Ava were removed from the network, those four individuals would be entirely disconnected from the other employees. Thus, Ava and Ethan have high betweenness centrality. In contrast, Dmitri and others on the edge of the network have a betweenness centrality of 0. Even Gabe, who has a degree of 5 and is in the center of the graph, has a relatively low betweenness centrality (6.5) because so many of his edges connect people who are already connected through others. In NodeXL, betweenness centrality scores are doubled for directed networks, though the “shortest paths” do not consider directionality in the calculation. Closeness centrality Another characteristic you may care about is how close each person is to the other people in the network. If information needed to flow through the network, some people would be able to get a message to all the other people relatively quickly (i.e., in few steps), whereas others may require many steps. Closeness centrality is a measure of the average shortest distance from each vertex to each other vertex. Specifically, it is the inverse of the average shortest distance between the vertex and all other vertices in the network. The formula is 1/(average distance to all other vertices). The inverse is used so that a higher closeness centrality indicates a more desirable centrality score (i.e., a shorter average distance to other vertices). For example, in the ABCD network, Ethan has the highest closeness centrality score, because he sits right in the “middle” of the network—not too far away from those in the top half of the network and not too far away from those in the bottom half of the network. In contrast, Dmitry, Liu, Camila, and Ben have the lowest closeness since they are so far removed from the majority of the other vertices. In NodeXL, closeness centrality assumes an undirected network, though it shows the same results for directed networks. Eigenvector centrality In many cases, a connection to a popular individual is more important than a connection to a lone individual.

II. NodeXL tutorial: Learning by doing

84

6.  Calculating and visualizing network metrics

The eigenvector centrality network metric takes into consideration not only how many connections a vertex has (i.e., its degree), but also the centrality of the vertices that it is connected to. Intuitively, it considers not just “how many people you know,” but also “who you know.” For example, in the ABCD network, Gabe has the highest eigenvector centrality (0.169) because his degree is relatively high (5), but also because those he connects to have high eigenvector centrality scores (e.g., Fay, Ji-yoo, Ethan, Hassan, and Ishita). In contrast, Ava has the same degree (i.e., number of connections) as Gabe, but those she connects with don’t have high eigenvector centrality scores since they have so few connections. As a result, Ava has a low eigenvector centrality score (0.043). In NodeXL, eigenvector centrality assumes an undirected network, though it shows the same results for directed networks.

cient measures how connected a vertex’s neighbors are to one another. More specifically, it is calculated as: (the number of edges connecting a vertex’s neighbors)/(the total number of possible edges between the vertex’s neighbors). For example, Ishita’s neighbors include Ethan, Gabe, and Ji-yoo. There are to edges connecting those three individuals (Ji-yoo to Gabe; Gabe to Ethan). However, there are three possible edges between them (those mentioned plus Ethan to Ji-yoo). This results in a clustering coefficient of 2/3 or 0.667 as shown in Figure 6.3. The value will always be between 0 and 1, since it is the percent of possible edges that are realized. It is the same formula as the overall network density, but only calculated on a subset of vertices.

PageRank The PageRank centrality metric is best known as the core metric behind Google’s search engine [3]. It is related to eigenvector centrality, but is designed for directed networks such as the world wide web. PageRank includes three distinct factors that determine the ultimate values for each vertex: (1) The number of vertices that link to the target, (2) the PageRank centrality of the linking vertices, and (3) the link propensity of the linking vertices. Consider a specific vertex representing a webpage called PageX. According to factor 1, the PageRank of PageX will increase if more vertices (i.e., websites) link toward it (i.e., it has a high in-degree). According to factor 2, the PageRank of PageX will increase if those who link to it have high PageRank themselves. This means that links are not all created equal. On the web, a link from cnn.com will increase a webpage’s PageRank far more than a link from a local blogger with a small following. According to factor 3, the PageRank of PageX will increase if those who link to it don’t link to many other vertices. In other words, links coming from “selective” linkers (those who only link to a small number of vertices) are more valuable than those coming from “frequent” linkers (those who link to a large number of vertices). In NodeXL, PageRank assumes a directed network, though it shows the same results for both directed and undirected networks. This network metric is not useful for the undirected ABCD network, but is useful for other networks such as the wikipedia page-to-page directed network (Chapter 14).

Network metrics must be interpreted differently depending upon the nature of the network. So far, we have examined a traditional network where the vertices represent people, and the edges represent direct connections between those people. However, many interesting networks connect people to things they are affiliated with (e.g., clubs, wiki pages they have edited, Facebook groups, classes). To better understand these “affiliation networks” and the meaning of network metrics associated with them, you will explore the Marvel Cinematic Universe affiliation network. Download the raw data in the file named Marvel_Movie_to_Character_Raw. xlsx file found at https://www.smrfoundation.org/ nodexl/teaching-with-nodexl/teaching-resources/. The network connects Marvel Universe movies to key characters that were in the movies. Data for the network was culled from the Marvel Cinematic Universe Wiki.2 Appearances in post-credit or deleted scenes are not included. As you will see, this bimodal network can be transformed into two unimodal networks: a characterto-character and a movie-to-movie network (see Advanced topic: Transforming a bimodal affiliation network into two unimodal networks).

Clustering coefficient In some cases, a person’s friends may be friends with each other. For example, Hassan's three friends Ji-yoo, Gabe, and Fay are all directly connected to one another, creating a clique. More generally, a clique or complete graph occurs when all vertices in a group are directly connected to each other. In other cases, a person’s friends may not be friends with one another. For example, none of Ava’s friends are connected to each other. The clustering coeffi-

6.4  Marvel cinematic universe network example

6.4.1  Visualizing and interpreting metrics in a bimodal network Start by looking at the available data on the Edges and Vertices worksheet. On the Edges worksheet, Vertex 1 includes the names of movies and Vertex 2 includes the names of key characters that appeared in those movies. The Vertex worksheet includes additional data for each vertex (see Figure 6.4) including the Type and corresponding Type_Code (1 is a character and 0 is a movie), Release_Date, Phase (a phase is a set of movies related to each other thematically and chronologically), 2 http://marvelcinematicuniverse.wikia.com.

II. NodeXL tutorial: Learning by doing

II. NodeXL tutorial: Learning by doing

FIGURE 6.4  Marvel Cinematic Universe network connecting Marvel movies to key characters that appear in those movies.

86

6.  Calculating and visualizing network metrics

IMDB_Score (average rating out of 10), Metascore (average rating out of 100), US_Opening (millions of dollars made in the opening weekend in the United States), Worldwide_Opening (millions of dollars made in the opening weekend worldwide), and URL (link to the IMDB page for the movie). Calculate the same Graph Metrics as you did for the ABCD network (Figure  6.1) since this network is also an undirected network.

A D VA N C E D T O P I C

Transforming a bimodal network into a person-to-person or affiliation-to-affiliation network typically requires the use of matrices or complex SQL queries. Some network packages, such as UCINET, will do this conversion for you [4]. Alternatively, it can be done for reasonably sized networks using Excel's built-in Pivot Table feature and SumProduct function. An example file called Marvel_Affiliation_Matric_Example.xlsx shows how to create the character-to-character and movie-tomovie networks from the bimodal Marvel Cinematic Universe dataset. It is available at https://www.smrfoundation.org/ nodexl/teaching-with-nodexl/teaching-resources/.

Transforming a bimodal affiliation network into two unimodal networks Bimodal affiliation networks like the Marvel Cinematic Universe network can be transformed into two single-mode networks: a person-to-person network (i.e., characterto-character network) and an affiliation-to-affiliation network (i.e., movie-to-movie network). The size of the new networks will depend on the number of people or affiliations. For example, in the Marvel network, there are 41 different characters, but only 17 movies. A person-to-person network created from affiliation data represents an indirect relationship between people. When the Marvel dataset is transformed into a character-tocharacter network, two characters will share an edge if they have been in a movie together. Furthermore, the edges will be weighted based on the number of movies they have been in together. For example, Iron Man and Pepper Potts have been in six movies together, so they have an edge weight of 6. You may call this the co-appearance network. Similar networks can be created from social media channels, such as a co-author network connecting Wikipedia editors who have co-authored the same pages (see Chapter  14); or a co-commenter network connecting YouTube commenters who commented on the same videos (see Chapter 13). Affiliation-to-affiliation networks also create weighted networks. When the Marvel dataset is transformed into a movieto-movie network, the weighted edges connecting movies are based on the number of shared characters in those movies. For example, Avengers: Infinity War shares 10 key characters with Captain America: Civil War. Other comparable networks connect Wikipedia pages based on the number of people who have edited both of them (see Chapter 14); or YouTube videos based on the number of people who have commented on both of them (see Chapter 13). The topic of the Wikipedia pages or YouTube videos may be completely different, but the people who contribute to them are the same. Thus, connections can link content together based on social structures, not direct linking between content. This type of inferred relationship is what serves as a basis for recommender systems such as Amazon's “Customers Who Bought This Item Also Bought” feature that relates books to other books based on the number of people who were “affiliated” (i.e., purchased) both books together.

Before interpreting the graph metrics, create a more informative network visualization, such as the one shown in Figure  6.5. Because this is a bimodal network, it is important to make movies and characters easily distinguishable. To address this, use the Autofill Columns feature with values shown in Figure 6.5. This will make movies solid squares, while characters remain disks. Additionally, set the color based on Phase (i.e., clusters of related movies), size based on eigenvector centrality, and add vertex labels (see Figure 6.5). After you reposition the vertices for graph readability, the bimodal network can be understood much better. Even with a clear visualization, examining the graph metrics can help highlight important vertices. You can sort on different network metrics to identify important characters or movies. Although the metrics are calculated in the same manner as they are calculated for unimodal networks, because it is a bimodal network, the interpretation is different. Degree has the most intuitive interpretation. For example, Ant-Man (see row 3 in Figure  6.4) has a degree of 3, which means he is in 3 movies. In contrast, Avengers: Infinity War has a degree of 24 (see row 7 of Figure 6.4), indicating that there were 24 key characters in that movie. Sorting on Degree is an easy way to examine which movies include the most key characters (Avengers: Infinity War; Captain America: Civil War; Avengers: Age of Ultron; Avengers), and which characters were in the most movies (Iron Man; Black Widow; Captain America; War Machine; Thor; Pepper Potts). Other metrics draw attention to different ways that movies or characters are important. As expected, Avengers: Infinity War has the highest betweenness, closeness, and eigenvector centrality given that it was designed to include all key characters from prior movies. However, these centrality metrics highlight unique positions in the network. For example, Ant-Man shows up with a very high betweenness centrality, because he is the only connection to the Ant-Man 1 and Ant-Man and the Wasp movies and their corresponding characters. Closeness centrality and eigenvector centrality both rank Iron Man and Black Widow

II. NodeXL tutorial: Learning by doing

II. NodeXL tutorial: Learning by doing

FIGURE  6.5  Completed Marvel Cinematic Universe network visualization with Phase mapped to color, Type_Code mapped to Vertex shape, eigenvector centrality mapped to Vertex size, and labels shown.

88

6.  Calculating and visualizing network metrics

FIGURE 6.6  Marvel movies mapping Degree to the X axis (logarithmic mapping), Betweenness Centrality to the Y axis (logarithmic mapping), Phase to color, and IMDB_Score to size (1.5–20). Axes and Legend are shown.

highly due to the fact that they are connected to so many of the key movies with many characters. Interestingly, several movies with six or more key characters (e.g., Black Panther, Guardians of the Galaxy, Thor: The Dark Underworld) have fairly low eigenvector and closeness centrality scores because they are connected to characters who do not show up in many other movies. Because this is an affiliation network, no movies directly connect to other movies, and no characters directly connect to other characters. As a result, the clustering coefficient is equal to 0 for all vertices.

6.4.2  Mapping graph metrics to X and Y coordinates In most layouts, the exact location of the vertices is not meaningful; only their position relative to one another has meaning. However, you may want to map network graph metrics, or other attribute data, to X and Y coordinates to visualize how two metrics interact with one another. Other metrics can be used to adjust visual properties, making it possible to display additional dimensions. For example, Figure  6.6 maps the movies onto the X and Y coordinates based on the degree and betweenness centrality respectively, using color and size to indicate Phase and IMDB score.

To recreate Figure 6.6, use the Autofill Columns feature. First, set the Vertex Label to Vertex so that the name of each character will be shown. Next set Vertex X to Degree, Vertex Y to Betweenness Centrality, making sure to check the box that says Use a logarithmic mapping in the Options for both metrics. Set the Vertex Size to IMDB_ Score (range from 1.5 to 20). Set Vertex Shape to Type_ Code similar to the prior graph, so movies show up as a square. To only show movies, and not characters, make the Vertex Visibility based on Type_Code with the Options as shown in Figure 6.6. Finally, navigate to the Edges worksheet and enter Hide into all of the Edge Visibility cells. This will hide all of the edges from the graph, making it far more readable. You can make the Legend and Axes visible using the Graph Elements drop-down menu in the NodeXL Ribbon. Try creating a similar network showing the characters by changing the Vertex Visibility options to display characters instead of movies.

6.5  CSCW 2018 conference Twitter network example The final example from this chapter will illustrate some of the metrics associated with directed networks that

II. NodeXL tutorial: Learning by doing



6.5  CSCW 2018 conference Twitter network example

i­nclude textual data. Specifically, you will be analyzing the network of Twitter users who posted at least one tweet that included “cscw” (an acronym that currently stands for a community of researchers studying Computer-Supported Cooperative Work and Social Media). Tweets were gathered from September 1, 2018 until November 23rd, 2018, which included time both before and after the 2018 CSCW conference,3 which occurred November 3–7 in New York City. Even though tweets were not gathered until September 1, some older tweets are included, since they were retweeted or replied to during the data collection timeframe. In Chapter 11, you will learn how to import your own Twitter networks. For now, you can download the file CSCW_2018_ Twitter_Raw.xlsx from https://www.smrfoundation.org/ nodexl/teaching-with-nodexl/teaching-resources/. The CSCW network is a good example of an EventGraph, or a “social media network diagram of conversations related to events, such as conferences” [5]. Such graphs can help make sense of the conversations around an event, helping to identify key individuals, subgroups, and general properties of the network compared to others. You will explore the graph metrics as part of this chapter and then use the same network in Chapter 7 to illustrate the value of filtering and grouping to bring clarity to a large network. The CSCW network is also a good example of a multiplex network (Chapter  3), since it includes three types of edges: Mentions, Replies to, and Tweet. After importing the network, browse through the Edges worksheet. Notice the Relationship column, which specifies the type of edge. If a user Mentions another Twitter user, then Vertex1 will include the sender and Vertex2 will include the user who was mentioned. If a user Replies to another user's tweet, then Vertex1 will include the sender and Vertex2 will include the person being replied to. These are directed edges that “point” from Vertex1 to Vertex2. Otherwise, if a person posts a Tweet that is not connected to any other tweets, the same username will show up in the Vertex1 and Vertex2 column. Graphically, this creates a self-loop, which is a loop that starts and ends at the user's vertex. Additional columns on the Edges worksheet indicate the text of the tweet (Tweet), Tweet Date (UTC), Imported ID (a unique identifier for each tweet), and Edge Weight. It is important to remember that each row in the Edges worksheet does not necessarily map to a single tweet. For example, if UserA mentions UserB and UserC in a single tweet, then two rows will be added to the Edges worksheet. The Imported ID can be used to count the total number of unique tweets, as discussed later. The Vertices worksheet includes a row for each Twitter user in the network, a profile image, and details about the Twitter user such as the number of Followers they have on Twitter as a whole. All of the extra data is pulled in from the Twitter API when using the Twitter importers 3 http://cscw.acm.org/2018/.

89

described in Chapter 11. Additional worksheets and metrics that have been calculated are explained below.

6.5.1  Calculating and interpreting directed network metrics In contrast to the small networks you have examined so far, many social media networks include hundreds or thousands of Vertices. In such cases, graph metrics become particularly important since initial visualizations obscure much of the data. Furthermore, network metrics can be used to filter out less important people as described in Chapter 7. The symbiotic relationship of network metrics and network visualization is extremely powerful, though it is often not used to its full potential. Because it can take a long time to calculate graph metrics for a network of this size, the file that you downloaded has already run the relevant metrics. Figure  6.7 shows the Graph Metrics settings that were chosen. Options dialog metrics are displayed to indicate specific settings that were added. For example, the Overall Metrics options dialog was used to add Relationship as an edge type (Figure 6.7). This in turn, creates new totals for each edge type (Mentions, Replies to, and Tweet) on the Overall Metrics worksheet (Figure 6.8) with counts of the number of edges for each. Notice the Overall Metrics also includes many metrics that we haven't yet seen. For example, it shows the Number of Edge Types as 3 (i.e., Mentions, Replies to, Tweets). It also shows the Reciprocated Vertex Pair Ratio and the Reciprocated Edge Ratio since it is a directed network. There are 278 self-loops (the same number as there are Tweet edges as expected). There are also many different connected components (106), most of which are single-vertex connected components (65). Because it is a directed network, all directed network metrics are chosen. Additionally, some undirected network metrics are chosen, such as eigenvector, betweenness, and closeness centrality. It is common for analysts to calculate such metrics, even for directed networks, but the interpretation of them is not exact. For example, betweenness centrality will still identify “bridge spanners,” but they may play that role because many disparate users mention them, or because they mention many disparate users. If you are looking for influencers, then identifying people with high In-Degree or PageRank, both of which are directed metrics, is more useful than identifying people with high Out-Degree or non-directed metrics such as Betweenness Centrality that may be driven by a person’s outbound links. Two metrics not yet examined include Edge Reciprocation and Vertex reciprocated vertex pair ratio. Notice on the Edges worksheet there is a column called Reciprocated?. If the value is Yes, then there is another edge that exists with the two users’ positions flipped. For example, there is a row where acm_cscw Mentions fcalefato. There is also a row where fcalefato

II. NodeXL tutorial: Learning by doing

II. NodeXL tutorial: Learning by doing

FIGURE 6.7  CSCW 2018 Network with directed graph metrics, time series, words and word pairs, and top items selected.



6.5  CSCW 2018 conference Twitter network example

91

their reciprocation percent is relatively low (0.129). This illustrates the importance of looking holistically at different metrics to fully understand a network.

6.5.2  Examining top items output When Top Items metrics were calculated, a new worksheet called Top Items was created (Figure  6.9).

FIGURE 6.8  CSCW 2018 Network Overall Metrics results.

Mentions acm_cscw. As a result, in both of those rows, the Reciprocated? column shows a Yes. A related metric on the Vertices worksheet is shown in the Reciprocated column. This shows the percent of vertex pairs that are reciprocated. It helps identify individuals who are primarily involved in conversations, since the people they reply to or mention also reply to or mention them. For example, some users such as farbandish (0.366), niloufar_s (0.361), and morganklauss (0.409) all had high values because they were actively participating in conversations. Not surprisingly, they also had both high In-Degree and Out-Degree. In contrast, individuals such as katestarbird (0.005) and snaglee2401 (0.027) have low reciprocated edges, in this case because they were mentioned by many users (i.e., had high In-Degree), but mentioned or replied to relatively few (i.e., had low Out-Degree). In general, the more popular someone is, the more difficult it is to have high reciprocation scores. A good example of this is the acm_cscw account, which worked hard to mention and reply to 77 different users (i.e., Out-Degree is 77). However, because they were mentioned or replied to 380 different times,

FIGURE 6.9  Top Items worksheet showing the top 10 users based on Followers, In-Degree, and PageRank in the CSCW 2018 Network.

II. NodeXL tutorial: Learning by doing

92

6.  Calculating and visualizing network metrics

This includes the metrics that were indicated in the Top Item Metrics Options dialog (Figure 6.7). The first list indicates the users in the network with the highest number of overall Twitter Followers. These can be thought of as global influencers. The second and third list show the top 10 individuals based on In-Degree and PageRank, which help identify the local influencers, or people that the CSCW network is frequently mentioning and retweeting. This includes the official cscw account (acm_cscw), researchers (e.g., katestarbird, informor), academic departments (vt_cs), and topics of discussion (warcraft, the account for the game World of Warcraft which was presented on). Additional lists can be created in the options dialog (Figure  6.7) for items such as the most common hashtags, URLs, words, people replied to or mentioned, or most active tweeters.

6.5.3  Examining time series output When Time Series metrics were calculated, the results were put into a new Time Series worksheet that includes a pivot table and associated graph (Figure  6.10). The graph shows the total number of unique tweets after they have been bucketed into days (since Days was chosen in the Time Series option box shown in Figure 6.7). Remember the total number of unique tweets is not the same as the total number of edges, since edges are ­often

duplicates (e.g., if user1 mentions user2 and user3 in the same tweet, then two edges are created). However, since the Unique edges by this column was set to Imported ID (Figure 6.7), the graph and corresponding data represent unique tweet counts as desired. Since Add a slicer for was set to the Relationship column (Figure  6.7), a filtering box is displayed. It allows you to filter based on Mentions, Replies to, and Tweets (Figure 6.10). If you click on one of the types of Relationship, it will filter the graph to only show tweets of that type. You can also choose multiple types. You can add different slicers (Figure 6.7) to examine other factors, such as location or time zone.

A D VA N C E D T O P I C Working with textual data

NodeXL includes the ability to perform text analysis, such as sentiment analysis (see Chapter 8 for a more detailed explanation). In Graph Metrics, you can choose Words and word pairs, and use the Options dialog to set up the type of analysis you desire (e.g., see Figure 6.7). For example, in the CSCW Network, a Sentiment Analysis was performed based on the default NodeXL settings. Sentiment analysis examines text to identify how “positive” or “negative” messages are. For example, positive messages may

FIGURE 6.10  Time Series worksheet showing a graph and pivot table grouped into Days with an additional slicer that allows you to filter what is displayed based on the Relationship column (Mentioned, Replies to, Tweet).

II. NodeXL tutorial: Learning by doing



93

References

include words like “abundant,” “keen,” “lawful,” and “super.” Negative messages may include words like “abuse,” “mediocre,” “wrong,” and “rant.” Many frequent words, such as “a,” “the,” “ever,” can be skipped from the analysis (Figure 6.7). For the CSCW Network, the Tweet message content was used. However, you could run it on other data, such as emails, YouTube comments, Wikipedia page content, etc. Additionally, this tool can be used to measure things other than sentiment. Simply replace the words in each of the three categories on the Options page with other sets of words. For example, words could be classified into buckets that indicate different product lines or companies by adding words such as the names of Nike shoes vs the names of Adidas shoes. Running these metrics creates a new worksheet called Words and another called Word Pairs. The Words worksheet includes all words that occur more than once (if indicated in the Options), along with the number of times they occur (Count), the Salience (a measure of how often the word occurs compared to other words), and a TRUE or FALSE statement indicating if the word is in one of the predetermined lists in the Options dialog (i.e., positive or negative word lists). The top includes summary data for all Positive, Negative, and Non-categorized words. The Word Pairs worksheet includes similar data, but for pairs of words that show up in the same message (e.g., tweet). It also includes a Mutual Information column that indicates the strength of the connection between the two terms. In addition to creating the new worksheets, the Word and Word Pairs feature also adds new columns to the far right of the Edges and Vertices worksheets. On the Edges worksheet, the new columns identify the number and percentage of words in each category (i.e., positive, negative, non-categorized), as well as a total word count (Edge Content Word Count). They are only calculated for unique messages (i.e., tweets), which explains why many values are blank in the CSCW Network. The Vertices worksheet reports similar content, but it is based on all messages associated with a user. Similar data can also be calculated for each group (Chapter 7). A related, but separate metric, is the Edge creation by shared content similarity feature. This feature allows you to create edges based on the similarity of the textual content used by two vertices. Instead of direct connections such as Mentions, these edges connect people who use similar words. Explanations for the use of this tool and details of the calculations are provided in the explanatory section of the Graph Metrics dialog.

6.6  Practitioner’s summary Social network analysis provides a set of powerful quantitative graph metrics for understanding networks and the individuals and groups within them. These

i­nclude aggregate network metrics such as graph density, diameter, reciprocated vertex pair ratio, and number of connected components, which characterize the network as a whole. They also include vertex metrics related to networks such as degree, in-degree, out-degree, betweenness centrality, eigenvector centrality, closeness centrality, PageRank, and clustering coefficient that can be used to identify unique or important people within a network. These metrics can be mapped onto visual properties such as size and opacity to help more easily make sense of the data, as was shown for the ABCD network. Affiliation networks, such as the Marvel Cinematic Universe network connecting movies-to-characters, have unique properties and their metrics must be interpreted carefully. Visualizations can combine calculated metrics (e.g., degree, betweenness centrality) with other attribute data (e.g., movie ratings; opening weekend proceeds) to gain insights into networks. NodeXL also provides text analysis features, time series analyses, and identifies top items when working with rich datasets such as the CSCW Twitter Network.

6.7  Researcher’s agenda The network metrics in NodeXL are widely used because they reveal important properties of individuals in a network [6–10]. However, their computation can be slow, so research efforts on improved algorithms (e.g., [11]), parallelization of execution using multiple processors, and the use of specialized graphic co-processors to speed computation are important. Improved centrality metrics for different types of graphs, such as bimodal and weighted graphs are also being actively explored (see [10] for initial attempts at some of these). The combination of natural language processing (i.e., text analysis) and social network analysis is providing promising results (e.g., [12]). Other metrics are regularly being created to help discover important vertices, edges, motifs, cycles, and other structural features, such as triangles, cliques, near-cliques, chains, holes, and more. Some are specific to certain platforms (e.g., Twitter [13]), while others are more generic.

References [1] M.E.J.  Newman, M.  Girvan, Finding and evaluating community structure in networks, Phys. Rev. E 69 (2004) 026113. [2] M.E.J.  Newman, Mathematics of networks, in: L.E.  Blume, S.N. Durlauf (Eds.), The New Palgrave Encyclopedia of Economics, second ed., Palgrave Macmillan, Basingstoke, 2008. [3] L. Page, S. Brin, R. Motwani, T. Winograd, The PageRank Citation Ranking: Bringing Order to the Web, Stanford InfoLab, 1999. [4] R.  Hanneman, M.  Riddle, Chapter  17: Two Mode Networks. Introduction to Social Network Methods, University of California, Riverside, CA, 2005. Riverside (published in digital form at http://faculty.ucr.edu/~hanneman.

II. NodeXL tutorial: Learning by doing

94

6.  Calculating and visualizing network metrics

[5] D.L.  Hansen, M.A.  Smith, B.  Shneiderman, EventGraphs: charting collections of conference connections, in: 2011 44th Hawaii International Conference on System Sciences, 2011, pp. 1–10. [6] P.  Bonacich, Power and centrality: a family of measures, Am. J. Sociol. 92 (5) (1987) 1170–1182. [7] L.C. Freeman, A set of measures of centrality based on betweenness, Sociometry 40 (1977) 35–41. [8] L.C. Freeman, Centrality in social networks: conceptual clarification, Soc. Networks 1 (3) (1979) 215–239. [9] D. Koschützki, K.A. Lehmann, L. Peeters, S. Richter, D. TenfeldePodehl, O. Zlotowski, Centrality indices, in: U. Brandes, T. Erlebach (Eds.), Network Analysis: Methodological Foundations, SpringerVerlag, 2005, pp. 16–61. LNCS 3418.

[10] S.  Wasserman, K.  Faust, Social Network Analysis: Methods and Applications (Structural Analysis in the Social Sciences), Chapter 5, Cambridge University Press, Cambridge, UK, 1994. [11] M.  Riondato, E.M.  Kornaropoulos, Fast approximation of betweenness centrality through sampling, Data Min. Knowl. Disc. 30 (2) (2016) 438–475. [12] A.  Bermingham, M.  Conway, L.  McInerney, N.  O'Hare, A.F. Smeaton, Combining social network analysis and sentiment analysis to explore the potential for online radicalisation. In Social Network Analysis and Mining2009, ASONAM'09 (2009) 231–236. [13] S.  Bruns, S.  Stieglitz, Towards more systematic Twitter analysis: metrics for tweeting activities, Int. J. Social Res. Methodol. 16 (2) (2013) 91–108.

II. NodeXL tutorial: Learning by doing

C H A P T E R

7 Grouping and filtering O U T L I N E 7.1 Introduction

95

7.2 U.S. Senate voting analysis 7.2.1 Filtering edges to identify groups within a network 7.2.2 Using dynamic filters 7.2.3 Creating groups based on vertex attribute

96 96 98 99

7.3 CSCW 2018 Twitter network analysis 101 7.3.1 Filtering out self-loops using the edge visibility column 101 7.3.2 Grouping and visualizing connected components 103 7.3.3 Using dynamic filters to filter based on time 103 7.3.4 Filtering based on vertex metrics and the visibility column 105

7.1 Introduction Most real-world social media networks are large and messy, much like the CSCW 2018 Network you began to examine last chapter. Visualizing and making sense of large networks can be challenging, particularly if they are densely connected. In this chapter, you will learn several different strategies for analyzing and visualizing large network datasets. One strategy for understanding large networks is to filter out information. Many criteria can be used to filter out vertices and/or edges. For example, vertices that have low centrality scores can be filtered out, leaving in only those who are most important in the network. Other data associated with vertices, such as age, country of origin, time zone, or number of Twitter followers can be used to filter vertices. In this chapter you will filter out vertices to gain insights into the most important individuals in the CSCW 2018 Twitter network. Filtering can also be applied to edges. For example, if edges represent the number of email messages exchanged between two people, a network of

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00007-8

7.3.5 Automatically identifying groups based on network clustering algorithms 7.3.6 Group properties and metrics 7.3.7 Group layout and labels 7.3.8 Creating subgraph images

106 107 107 109

7.4 Federal communications commission (FCC) lobbying coalition network

109

7.5 Practitioner’s summary

112

7.6 Researcher’s agenda

112

References

112

Additional resources

113

“strongly” connected individuals may filter out those who have sent less than 10 messages to one another (see Chapter  9). In this chapter you will filter edges based on co-voting percentages between U.S. senators in the 115th Congress (2017–2018). Finally, filtering can also be an excellent method for exploring networks, particularly when using dynamic filtering tools such as those provided in NodeXL. Another strategy for understanding large networks is examining groups of vertices. Many large networks are a complex combination of smaller groups or subgraphs. High school networks in the United States consist of subgroups of jocks, nerds, goths, and the like. Facebook networks are made up of clusters of family members, schoolmates, work colleagues, and other forms of association. Legislative bodies like the U.S. Congress contain two main political parties and numerous smaller coalitions. Identifying groups within a network and mapping their relationship to one another can be essential to making intelligent strategic decisions. Network analysis can help identify competing or complementary groups, potential allies to form a

95

© 2020 Elsevier Inc. All rights reserved.

96

7.  Grouping and filtering

FIGURE  7.1  A network of three densely connected clusters (i.e., groups), each shown inside a dashed circle. Ties between clusters are rare and less dense.

­ owerful group, and individuals who can connect you p to a new group. Social network analysis provides a set of tools for identifying and understanding groups, also called clusters or communities by network researchers. In the language of network analysis, clusters are pockets of densely connected vertices that are only sparsely connected to other pockets [1]. For example, Figure  7.1 shows a network consisting of three densely connected pockets (within the dotted circles) that are loosely connected to each other by only a few ties. One way to create groups is to associate vertices that have a shared attribute (e.g., people from California vs those from Maryland). However, often the most interesting groups are those that emerge from network connections, not formal group membership. Several algorithms exist that create these organic clusters based solely on network ties. For example, an analysis of a corporate email network (Chapter  9) can provide an authentic grouping of individuals based on communication patterns rather than formal reporting hierarchies. In this chapter you will learn to identify and visualize groups from several networks.

7.2  U.S. Senate voting analysis In this section you will analyze the voting patterns of U.S. senators, identifying clusters of senators connected together based on similar voting patterns. Chris Wilson of Slate magazine provided the original voting network data from 2007, which inspired us to create up-to-date versions for each of the 110– 115th Congresses, covering the years 2007 through 2018. All datasets are available on the book website. In this chapter, we will use the 115th Congress data, which covered the time period from January 4, 2017

to December 21, 2018 (the last votes in the dataset when we pulled it) and included a total of 599 roll call votes. You can download the source file named Senate115.xlsx at https://www.smrfoundation.org/ nodexl/teaching-with-nodexl/teaching-resources/. The source data was gathered from https://voteview. com [2]. The original data is not in a network format. Instead, the original data files include information on how each member of the Senate voted on each roll call. We transformed the data into a ­co-voting ­n etwork as described below. The Senate co-voting network is created from data that connects senators to one another based on the number of times they vote the same way (i.e., both voted yea or both voted nay on a bill; we do not draw an edge if they both abstained or were absent). For example, senators Alexander (R, Tennessee) and Baldwin (D, Wisconsin) voted the same only 226 times (39.5% of the time), whereas Senator Alexander (R, Tennessee) and Barrasso (R, Wyoming) voted the same 538 times (94.1% of the time) (Figure  7.2). Clearly the two Republicans have a stronger connection to one another than the Democrat and Republican senators. The network is undirected and is weighted based on the percent of similar votes (see the Voted_ Same column in Figure  7.2). Using the raw number of similar votes can be problematic in cases where senators were frequently absent for votes (i.e., were campaigning). For this reason, the dataset also includes the total number of votes cast by each senator (Vertex1_Total & Vertex2_Total) and the percentage agreement (Percent_Agreement), which is calculated using the lowest of the two senators’ total votes as the denominator. The Vertices worksheet includes data about each senator including the senator’s party affiliation, the state the senator represents, the total number of votes the senator cast, and their unique ICPSR number (i.e., a unique identifier created by the Inter-university Consortium for Political and Social Research for each congressman) [2].

7.2.1  Filtering edges to identify groups within a network When working with weighted networks such as the 115th Senate network, it is often necessary to filter out some of the edges to identify subgroups. Because every senator voted the same as every other senator at least once, choosing Show Graph results in an uninformative dark mass of connections (Figure 7.2). To make the graph more meaningful, you will want to only show edges between senators that have a high level of agreement. In other words, you want to filter out edges where the Percent_Agreement is underneath a certain threshold (e.g., 64%).

II.  NodeXL tutorial: Learning by doing



7.2  U.S. Senate voting analysis

97

FIGURE 7.2  Unfiltered 2017–18 U.S. 115th Senate co-voting network showing all senators connected to one another. Other columns in the NodeXL Edges worksheet show the number of times each pair of senators voted the same and their percent agreement. “Raw” visualizations like this require refinement to display useful insights.

A D VA N C E D T O P I C Visibility column options

Like other visual attribute columns, the Visibility column can be filled using Autofill Columns, populated using formulas, or manually by typing in the desired option. The following options are available: • Show if in an edge. If the vertex is connected to anything else via an edge, show it. Otherwise, ignore the vertex row. This is the default setting. • Skip. Skip the vertex row and any of its edges. Or, if on the Edges worksheet, skip the edge row that is selected. Do not read them into the graph. This essentially pretends the data is not in the spreadsheet. For example, if you choose to calculate Graph Metrics, all skipped vertices are excluded from the

calculations. They will also not be part of Groups and will not affect the layout of other vertices. • Hide. Hide the vertex and its edges from showing up in the graph pane. Or, if on the Edges worksheet, skip the specific edge that is selected. This is what the dynamic filters do. Unlike skipped vertices, hidden vertices and edges are included when network graph metrics are calculated and they affect the layout of all vertices in the graph pane, even if you can’t see them. • Show. Show the vertex regardless of whether it is part of an edge. Or, if on the Edges worksheet, show the specific edge that is selected.

II.  NodeXL tutorial: Learning by doing

98

7.  Grouping and filtering

Begin by using the AutoFill Columns feature to set the Vertex Label to the Vertex column (i.e., the senator’s last name). Set the Vertex Tooltip to State. Then navigate to the Shape column on the Vertices worksheet and change the values for all vertices to Label. This will make all senator’s names clearly visible and allow you to see their State on mouseover. Use the Edges tab on the Autofill Columns feature to set the Edge Opacity and Edge Visibility to Percent_Agreement. You’ll need to use the Options for each of them. Set the Edge Visibility Options so that the edge only shows up If the source column number is: Greater than 0.64 (Figure  7.3). Using the default settings, this will only show edges between two people who have a greater agreement level than 0.64, which is the average percentage agreement between all pairs of senators. See the Advanced topic: Visibility column options for more details. Set the Edge Opacity Options to those described in Figure  7.3. Finally,

change the Color of the edges on the Edges worksheet to something more distinct than gray (e.g., 128, 128, 192). After refreshing the graph and positioning the vertices (see Section 4.4.1), the result should look like Figure  7.3. The largely two-party system in the U.S. Senate becomes very apparent, with a cluster of conservative senators and a cluster of liberal senators, and a few senators in the middle.

7.2.2  Using dynamic filters NodeXL allows you to dynamically filter out edges or vertices based on any data fields on the Edges and Vertices worksheets. This can be an excellent way of exploring a dataset, without making any permanent decisions about how to display it. Click on the Dynamic Filters button above the graph (see Figure  7.4) and you will be presented with a new window that lets you set the minimum and maximum values for each variable. Find the frequency table and sliders associated

FIGURE 7.3  Filtered 2017–18 U.S. 115th Senate co-voting network after using the NodeXL Autofill Columns window with Edge Visibility Options set above 0.64 and Edge Opacity Options set to a range of 0.64 (edge opacity 10) to the largest number in the column (edge opacity 100). Labels are set to the Vertex column and the Vertex Shape is set to Label.

II.  NodeXL tutorial: Learning by doing



7.2  U.S. Senate voting analysis

99

FIGURE 7.4  Dynamic Filters window used to “hide” edges with Percent_Agreement below 0.95. Filter opacity is set to 5% in order to faintly show the hidden connections.

with Percent_Agreement. Notice from the graphic that there is a bimodal distribution (i.e., a group of senators with low agreement toward the middle-left side and a group of senators with high agreement on the right-hand side). Use the sliders associated with the Percent_Agreement data to explore the dataset. For example, drag the left slider to the right of the current visibility threshold (0.64) to exclude any edges below, say 0.95 (i.e., 95% agreement). This will hide the edges and vertices from being displayed, although they still will affect the layout. You can faintly show the hidden vertices by setting the Filter Opacity to 5% as shown in Figure 7.4. This type of dynamic analysis can help identify differences between the political parties, as well as sub cliques within them. For example, Figure 7.4 makes clear that the Republicans (the top cluster) voted as a block more often than the Democrats (the bottom cluster). Comparing the different Senate datasets available on the book website can reveal trends over time. Once you are done with your explorations, you can revert to the original image by clicking on Reset Filters and close out of the dialog.

7.2.3  Creating groups based on vertex attribute Sometimes you will have data that describes attributes of the people in your network. For example, our Senate dataset includes information on the political party of each senator (see Party column on the Vertices worksheet). Values include D (Democrat), R (Republican), and I (Independent). To visually display this information, click on the Groups drop-down menu in the NodeXL ribbon and choose Group by Vertex Attribute. Choose Party from the first drop-down menu, since this is the attribute you want to group based upon. Then choose Categories from the second drop-down menu, since the data is categorical in nature. Notice, though, that groups can be created based on numerical data or date and time data. When you click OK, NodeXL will create two new worksheets. The Group Vertices worksheet includes a table showing each unique Vertex, alongside its Group and unique Vertex ID as shown in Figure 7.5. Additionally, the Groups worksheet is created, which includes one row for each unique group, alongside information on how to visually display the group, labels, and metrics, as shown in the left-hand side of Figure 7.6.

II.  NodeXL tutorial: Learning by doing

FIGURE 7.5  The Group Vertices worksheet that maps each vertex to exactly one cluster and a unique Vertex ID.

FIGURE 7.6  The Senate co-voting network after applying the Group by Vertex Attribute based on political Party. Notice that the labels disappeared because the Shape information is now pulled from the Groups worksheet. The Vertex Colors were automatically populated and are not good choices for this context.



7.3  CSCW 2018 Twitter network analysis

101

FIGURE  7.7  The Senate co-voting network after using the Group Option dialog to pull Vertex Color from the Groups worksheet and the Vertex Shape from the Vertices worksheet.

The image in Figure 7.6 is problematic in a couple of ways, which can be fixed. You will noticed that the vertices that are part of each group have been assigned different colors. This is good. However, the automatically assigned colors are not appropriate for this context, where red is typically associated with Republicans and blue is typically associated with Democrats. This can be easily fixed by changing the Vertex Color values next to R, D, and I on the Groups worksheet to be Red, Blue, Green (see Figure 7.7). Second, the Vertex Shape is now set to Disk (on the Groups worksheet), instead of Label (on the Vertices worksheet). In order to fix this, open the Group Options dialog (via the Groups dropdown menu in the NodeXL ribbon) and choose the settings shown in Figure  7.7. The result is a visualization that clearly shows each senator, their political party, and their network position.

7.3  CSCW 2018 Twitter network analysis An analysis of the CSCW 2018 Twitter network, introduced in Chapter  6, will help introduce several ­additional features of NodeXL that relate to filtering and grouping. Like most networks pulled from social media, the original network provides few insights without further refinement (see Figure 7.8). In this section, you will learn how to bring clarity and understanding to this type of real-world network.

7.3.1  Filtering out self-loops using the edge visibility column As described in Chapter  6, Twitter networks include three types of edges in the Relationship column: Mentions, Replies to, and Tweet. In order to focus in on the relationships between people, you can Skip the ­messages

II.  NodeXL tutorial: Learning by doing

102

7.  Grouping and filtering

FIGURE 7.8  CSCW 2018 Twitter network before it is filtered and grouped. The formula shown in the top bar sets the Visibility column to Skip if it is a Tweet, and Show if it is a Mentions or Replies To.

that only connect to one person—namely the Tweet messages. These are visually displayed as an edge that circles back and points at the same vertex it originated at. Because the data in the Relationship column is not numerical, you cannot use Autofill Columns to filter out the edges. Instead, write the formula shown in Figure 7.8 to set the Visibility column to Skip if the message type is

Tweet, or, if not, set it to Show. After you Refresh Graph, there should not be any self-loops remaining. This method is preferred over using Excel's built-in filtering tool (e.g., accessible via the drop-down menu in the Relationship column header). While using the filtering tool will not necessarily break anything, it will hide the rows, which may cause inadvertent mistakes to be made later.

A D VA N C E D T O P I C

Count and merge duplicate edges Many networks, similar to the CSCW 2018 Twitter network, have duplicate edges. For example, as shown in Figure 7.8, dh_30x replied to w_cscw two times (see rows 18 and 19). A view of the Overall Metrics worksheet reveals that there are 1593 duplicate edges in the CSCW 2018 Twitter network. This can be useful in many contexts. For example, in this case, the content of the tweet is preserved. Unfortunately, it can cause some problems. For example, you may want to visualize the number of tweets as a weighted line. However, to do this, you will need to count how many

duplicates there are for each row and represent that in a column (e.g., Edge Weight). Sometimes there is no need to preserve each individual row, and they can be merged into a single unique row, to simplify the dataset. NodeXL allows you to do these things by using the Prepare Data dropdown menu found in the NodeXL Ribbon and choosing the Count and Merge Duplicate Edges option. This will open a new dialog like the one shown in Figure 7.9. The first option indicates that after the duplicate edges are merged, a new Edge Weight column will

II.  NodeXL tutorial: Learning by doing



7.3  CSCW 2018 Twitter network analysis

FIGURE 7.9  Count and Merge Duplicate Edges dialog with settings that will merge all duplicate edges (i.e., rows) that contain the same values in the Vertex 1, Vertex 2, and Relationship columns.

be added to the Edges worksheet and populated with the number of edges that were combined. For example, the two rows where dlh_30x replied to w_cscw would have an Edge Weight of 2 appear. The second option will collapse the duplicate edges into a single row. For example, if you were to check the Merge duplicate edges box, the number of rows in the Edges worksheet would be reduced to the number of

7.3.2  Grouping and visualizing connected components Perhaps the simplest method to identify groups within a network is to group together vertices that are part of different connected components. Find the groups connected vertices that are completely separated out from other groups of connected vertices. To do this, choose Group by Connected Component from the Groups dropdown menu. This will populate the Groups worksheet with a different row for each connected component in the graph. They will be sorted in order from largest to smallest. In many cases, there will be one large connected component and many smaller components. To make sure they do not overlap one another, open the Layout Options dialog (found in the Layout dropdown menu above the graph), and check the box shown in Figure 7.10 that lays out the smaller components in boxes at the bottom of the graph pane. You can customize the maximum size of the connected components (10) and size of the invisible boxes (25). You may want to also update the Fruchterman-Reingold layout settings to those shown in Figure 7.10 to achieve a better layout. The graph showing the entire network in Figure 7.10 provides a good sense of the overall scope and structure of the network. In this case, there seem to be many people who are connected directly or indirectly in the large blue connected component. There are also a number of smaller connected components that have not mentioned or replied to anyone in the larger network. It is also apparent that there seems to

103

unique edges. In our example, the two dlh_30x replied to w_cscw rows would be collapsed into a single row. In this network, you may NOT want to do this, since the tweet content of one of the collapsed edges would be permanently erased. However, this feature can come in handy in many instances. The section titled Columns that determine whether edges are duplicates allows you to account for cases where you have different edge types, such as the CSCW 2018 Twitter network. If you were to select the first option, then all edges from one person to another would be counted or merged, even if they were of different types (e.g., replies to, mentions, and tweet). This may be desirable, but in many cases it is not. The example in Figure  7.9 shows the Relationship column to also determine uniqueness. By checking this box, the counts and/ or merged edges do NOT roll up Replies to and Mentions (values in the Relationship column). For example, even if you merge edges, there will be two rows that show caylery in Vertex1 and farbandish in Vertex2. One would be of Relationship type Mentions (with an Edge Weight of 2) and the other would be of Relationship type Replies to (with an Edge Weight of 3).

be a core group of people in the center of the large component, as well as a few key people that many people mention or reply to (e.g., the individual toward the right-hand side with a fan of individuals pointing toward her).

7.3.3  Using dynamic filters to filter based on time One particularly useful way to use Dynamic Filters is to examine changes in a network over time. Since Twitter data has timestamps on the tweets that were sent, you can use the sliders to “play back” the discussion over time. Open Dynamic Filters and look at the first two Edge Filters called Relationship Date (UTC) and Tweet Date (UTC)(see Figure 7.11). For this network, these columns are identical, so you can use either one. Update the values in the Relationship Date (UTC) fields so the range covers only the time period of the conference as shown in Figure 7.11. Choose the Lay Out Visible Vertices Again option from the Lay Out Again drop-down menu as shown in Figure 7.11. This will make it so that the hidden vertices do not impact the layout of those that are shown. You can explore additional ways of playing with the Relationship Date (UTC) fields. Reset the filters back to their starting point by clicking the Reset Filters button (Figure 7.11). Then drag the right-hand slider all the way to the left and slowly slide it to the right. This will show a dynamic view of the network as it unfolded over time. A more precise way of doing this

II.  NodeXL tutorial: Learning by doing

FIGURE 7.10  CSCW 2018 Twitter network after grouping by Connected Component and laying out the smaller connected components in boxes across the bottom.

FIGURE 7.11  Dynamic Filter window with the Relationship Date (UTC) set to show only tweets that occurred during the conference dates of Nov. 3, 2018 through Nov. 7, 2018. The Lay Out Visible Vertices Again feature is used so the hidden vertices don’t affect the layout of the visible vertices.

II.  NodeXL tutorial: Learning by doing



7.3  CSCW 2018 Twitter network analysis

is to click on the day portion of the right-hand side date value and use the up or down arrow keys to increase or decrease the day. You will likely notice significant bursts of activity, such as a post by katestarbird on Sep. 4, 2018 that was widely retweeted. Once you are done exploring, click the Reset Filters button (Figure 7.11) and close out.

7.3.4  Filtering based on vertex metrics and the visibility column While the full network visualization is a nice overview, it is hard to focus in on groups within the large component or identify individuals who are important within it. For that analysis, it is necessary to filter out the less important vertices as determined by network metrics. Navigate to the Metrics columns on the Vertices worksheet and use the Sort Largest to Smallest feature (e.g., on the In-Degree column title). This can help you identify important individuals, as well as cut-off points for filtering. For example, after sorting on In-Degree, it becomes apparent that there are some outliers, such as the acm_cscw account that has an in-degree of 380 (i.e., 380 unique individuals who have mentioned or replied to them). You may also notice that the majority of user accounts have an

105

in-degree of 0, 1, and 2. This power-law distribution is typical of real world social networks. Because in-degree is a useful measure of local importance within a particular network, you can use it to filter out those with a low score. Use Autofill Columns to skip the Visibility of vertices with an In-Degree of 2 or less as shown in the Figure  7.12 Vertex Visibility Options dialog. You can also set the Size of the vertices to be based on In-Degree (checking the Use a logarithmic mapping) and set the Vertex Tooltip to be based on the Vertex column (i.e., where usernames are recorded). The resulting image shown in Figure 7.12 is a bit less overwhelming than the unfiltered version, but is still somewhat cluttered. This is largely because of the prominent position of the acm_cscw user, whose connections are overwhelming some of the other structures, making them less apparent. Try removing the user by navigating to the Vertices worksheet and manually choosing Skip in the Visibility column on the acm_cscw row as shown in Figure 7.13. This will remove acm_cscw from the graph, though you should be careful since running Autofill Columns again will overwrite this manual change. The resulting graph (see Figure 7.13) helps you realize that the network is a bit more spread out than it might have otherwise appeared.

FIGURE 7.12  CSCW 2018 Twitter network after using Autofill Columns to filter out (i.e., Skip) vertices that are not greater than 2.

II.  NodeXL tutorial: Learning by doing

106

7.  Grouping and filtering

FIGURE 7.13  CSCW 2018 Twitter network after removing acm_cscw from the graph (right) by manually setting user’s Visibility to Skip (left).

7.3.5  Automatically identifying groups based on network clustering algorithms NodeXL can automatically identify groups within a network based solely on network structure. In contrast to the approach of using existing data about the attributes as used in Section  7.2.3, this approach is based solely on who is connected to whom. A number of different network “clustering” (also known as “community detection”) algorithms exist, which help find subgroups of highly inter-connected vertices within a network. NodeXL includes three such algorithms: Clauset-Newman-Moore, Wakita-Tsurumi [3], and Girvan-Newman (which can take a long time to run on large graphs). In all of these algorithms, the number of clusters is not predetermined; instead the algorithm dynamically determines the number it thinks is best. Each vertex is assigned to exactly one cluster, meaning that clusters do not overlap. The number of vertices in each cluster can vary significantly. In some cases, a single cluster can encompass all vertices, whereas in other cases, a cluster can consist of a single vertex. See Newman [4] for background on some of these and other community identifying algorithms. There is no “right” or “wrong” algorithm to use; instead, it is often useful to try out different ones and see which ones you believe provide the best results given your network. For example, in this network, the

­lauset-Newman-Moore algorithm results in fewer, C larger groups than the other algorithms, which provide more groups of a smaller size. Try applying the WakitaTsurumi clustering algorithm by clicking on the Groups dropdown menu in the NodeXL ribbon and choosing Group by Cluster and the checking the appropriate selector as shown in Figure 7.14. Notice that the data on the Groups worksheet is now updated to reflect the new groups.

A D VA N C E D T O P I C

Additional clustering algorithms There are a large and growing number of network clustering algorithms, also called community detection algorithms (see Newman and Girvan [1] for an overview). Many community clustering algorithms don’t scale well for large networks, forcing a tradeoff between quality and speed. Most algorithms, including the ones used by NodeXL, are based on undirected, unweighted networks. Although you can apply them to more complex networks and often get reasonable results, you may need to use a more specialized community detection tool that includes a range of algorithms that take into consideration the specific properties of your network data. Most tools will output data into a format that can easily be pasted into the Groups and Group Vertices worksheets, allowing you to take advantage of NodeXL’s rich visual features.

II.  NodeXL tutorial: Learning by doing



7.3  CSCW 2018 Twitter network analysis

107

FIGURE 7.14  CSCW 2018 Twitter network after applying the Wakita-Tsurumi clustering algorithm.

In addition, there are many non-network clustering algorithms that can operate on collections of vertex attributes [5]. These include k-means, hierarchical agglomerative, hierarchical divisive, and many more. For example, a non-­network clustering algorithm could be used to cluster people into groups that have similar participation patterns (e.g., those who use a certain collection of features similarly). This could then be represented on the network graph to see how those who use the system in a similar way are connected to one another.

7.3.6  Group properties and metrics The Groups worksheet includes many fields that can be useful in analyzing and visualizing networks. If you click on one of the Group worksheet rows, it will ­highlight all vertices associated with that row’s group. The Vertex Color and Vertex Shape columns have already been introduced. It is worth reminding you that you may need to use the Graph Options dialog to determine if vertex color and/or shape should be pulled from the Groups worksheet or the Vertices worksheet (see Figure 7.7). The Visibility column can be used to Show, Skip, or Hide each group. Choose Skip for the bottom 7 groups, which are all separated from the large component (see Figure 7.15). This will filter them out of the graph, and also make it so metrics won’t be

c­ alculated for them. Choosing Yes in the Collapsed? column lets you collapse an entire group into a single shape (whatever is specified in the Vertex Shape column) with a plus sign in the middle of it. The size of the shape depends on the number of vertices in the group. However, collapsing groups may hide important information, so keep them set to the default (i.e., No) for now. Calculate metrics for each group by opening the Graph Metrics dialog (see Chapter  6) and checking the box next to Group metrics (Figure 7.15). This will populate the Graph Metrics columns on the Groups worksheet, except for the groups that have Visibility set to Skip. Metrics include many of the metrics found on the Overall Metrics worksheet, such as the number of Vertices, Unique Edges, Total Edges, Graph Density, etc. Sorting on those columns can help identify key differences between the groups. For example, the group G5 has a very low Graph Density compared to most other groups, largely because there is one key person mentioning many other people.

7.3.7  Group layout and labels At times, it can be useful to visualize groups separately from one another. NodeXL allows you to do this using the the Layout Options… dialog as shown in Figure 7.16. Choose the option Lay out each of the graph’s

II.  NodeXL tutorial: Learning by doing

FIGURE 7.15  CSCW 2018 Twitter network Visual Properties and Graph Metrics columns after calculating the Group metrics using the Graph Metrics feature.

FIGURE 7.16  CSCW 2018 Twitter network using the Layout Options to display each group in its own box and hiding the edges between vertices in different groups. Group labels are also shown.



7.4  Federal Communications Commission (FCC) lobbying coalition network

groups in its own box and set the Intergroup edges: to Hide. The resulting graph, shown in Figure 7.16 is missing data (i.e., the hidden edge connections between each group), but this allows for a more focused comparative analysis of groups. To learn more about this novel group in a box layout technique, see Ref. [6]. To better identify important individuals within the graph, enter the names of important users (based on metrics) in the Labels column on the Vertices worksheet. Additionally, add group labels by entering the group name in the Labels column on the Groups worksheet (see Figure  7.16). To place the group labels in the upper-left corner, as shown in Figure  7.16, use the Graph Options dialog, choose the Other tab, choose Labels… and set the Position: to Top Left in the Group box labels section.

A D VA N C E D T O P I C Group by Motif

NodeXL is able to group by different network motifs, or in other words, different common visual patterns of connections. See Ref. [7] for a full description of this novel visualization technique. To access this feature, choose Group by Motif in the Groups drop-down menu to open the Group by Motif dialog box (Figure 7.17). As with all of the different grouping options, using this will overwrite all other groups that were previously calculated.

109

for each vertex in the network. Network scientists call these egocentric networks (see Chapter 3). They provide a personalized view of the network from the perspective of an individual vertex and are useful when comparing vertices to one another. For example, vertices with a similar structure (e.g., a hub and spoke) may play similar social roles such as a question answerer in a discussion forum (see Chapter 10). To create network subgraph images, click on the Subgraph Images button on the Analysis section of the NodeXL ribbon (top-right of Figure  7.18). The Subgraph Images dialog box will appear (Figure 7.18). The first option allows you to choose the levels of adjacent vertices to include in each subgraph. For example, the default of 1.5 will show edges connecting the source vertex with its direct neighbors, as well as any edges that connect the neighbors to one another. Choosing 2.0 will show all of those edges, plus edges connecting the source vertex’s neighbors with all of their neighbors. If the data were from a social networking site such as Facebook, a 2.0 setting would show your friends, which of your friends know one another, and all of your friends’ friends (FOAF). For now, replicate Figure 7.18 by using the default settings. This will generate a new column called Subgraph as shown in Figure 7.18. Subgraph images are positioned relative to each other based on the currently selected layout algorithm, so make sure you use an appropriate one (e.g., Fruchterman-Reingold) for your data. Subgraphs highlight important differences between vertices. To illustrate this point, compare the subgraph images of individuals. For example, clifflampe and asbruckman are part of a densely connected active discussion group. In contrast, stbridgetathena and warcraft are mentioned or replied to by people who are otherwise not connected to one another. While the Clustering Coefficient metric leads to similar insights, a visual representation using subgraphs can often illuminate more nuances, such as the fact that hypotext is always mentioned alongside someone else.

7.4  Federal Communications Commission (FCC) lobbying coalition network FIGURE  7.17  The Group by Motif dialog box that allows you to determine which type of network motif to collapse in the visualization.

7.3.8  Creating subgraph images Another useful way to understand complex networks is to view individual sections, or subgraphs, of the larger graph. NodeXL allows you to create Subgraph Images

The power of social network analysis and visualization is best realized when combining the approaches discussed so far (Chapters 4 through 7). In this section you will see a network visualization created in NodeXL by Pierre de Vries, a Research Fellow at the University of Washington’s Economic Policy Research Center. This example is based on his submission to the first annual Journal of Social Structure Visualization Symposium held in 2010.

II.  NodeXL tutorial: Learning by doing

110

7.  Grouping and filtering

FIGURE 7.18  Subgraph images dialog box with default settings and the resulting subgraph images inserted into the Subgraph column on the Vertices worksheet.

The social network shown in Figure 7.19 shows the relationship between organizations that lobbied the FCC on just one of the hundreds of issues that are before it: Docket 01-92 on intercarrier compensation. This proceeding is a battle over the fees that telephone companies pay each other when phone calls move between them. Legislation requires that most interactions between organizations and the FCC are publicly documented. This network data was extracted from metadata about these filings reported via the FCC’s Electronic Comment Filing System. The network captures all links between lobbying organizations during the duration of the proceeding (2001–2008). Vertices represent organizations that filed. Edges connect organizations that filed jointly, with edge

weight representing the number of joint filings. Darker edges reflect higher edge weights. Vertex size is proportional to the total number of filings that an organization made and is a proxy for lobbying investment. Well-funded companies and trade associations are prominent, although they are not necessarily well connected because they don’t need to be. Influence, measured by eigenvector centrality, is represented by the color of the nodes: the pinker the node, the better connected it is. Small companies can gain influence by linking different coalitions of local exchange carriers. Some organizations that hardly filed at all may be influential (i.e., small and pink), thanks to their many, straddling connections.

II.  NodeXL tutorial: Learning by doing



7.4  Federal Communications Commission (FCC) lobbying coalition network

111

FIGURE 7.19  Lobbying Coalition network connecting organizations (vertices) that have jointly filed comments on U.S. Federal Communications Commission policies (edges). Vertex size represents number of filings and color represents eigenvector centrality (pink = higher). Darker edges connect organizations with many joint filings. Vertices were originally positioned using Fruchterman-Reingold and handpositioned to respect clusters identified by NodeXL’s Find Clusters algorithm.

The Fruchterman-Reingold algorithm was used to prepare a preliminary network layout for these data. Next, the Group by Cluster feature was used to identify distinct clusters, which were used to guide the manual placement of vertices into visually intelligible positions. Once clusters were represented by their placement, the cluster colors were pulled from the Vertices worksheet instead of the Groups worksheet so that color could be used to represent eigenvector centrality. The file was exported as a high-resolution image, making it possible to zoom in on different sections of the graph and still read the vertex labels. Interviews with policy practitioners confirm that the graph clusters correspond to real-world coalitions and alliances. For example, the clusters in the top right and bottom right are rural telephone companies; a loose coalition of rural trade associations and

rural competitive local exchange carriers can be seen at the bottom middle of the illustration. The heavily connected cloud in the center of the chart shows competitive local exchange carriers who band together at various times in various permutations to make up strength in numbers. Because graph clusters and evolution represent ­real-world behavior, they can be used to improve public understanding and lobbying effectiveness. Insiders can use graphs to identify potential collaborators or defectors (e.g., by looking for coalition members who are bridges between groups). They can also use changes in connectedness to track the emergence or breakdown of consensus in a proceeding. Outsiders can grasp the overall structure and evolution of the proceeding without having to understand the entire record.

II.  NodeXL tutorial: Learning by doing

112

7.  Grouping and filtering

7.5  Practitioner’s summary Making sense of large networks can be challenging, particularly if they are densely connected. Several techniques used to simplify and clarify networks related to filtering. For example, edges can be filtered based on associated data as you saw in the Senate co-voting network (edges with a low co-voting percent were removed) and Twitter network (self-loops were removed). Vertices can also be filtered based on associated data or metrics, such as in-degree. Even groups can be filtered out using the Visibility column on the Groups worksheet. The Dynamic Filters feature allows you to interactively determine what is displayed on the graph by setting the position of sliders representing starting and ending ranges. These ranges can use data from graph metrics, timestamped data, or other attribute data associated with the network. Another technique for making sense of large, complex graphs is to break them into groups, which are also called clusters, communities, or subgraphs. Groups are pockets of densely connected vertices that are only sparsely connected to other pockets. Groups can be automatically identified using community detection algorithms such as the those behind NodeXL’s Group by Cluster feature (e.g., Ref. [3]). If group memberships are known (e.g., Republican and Democrat party affiliations), vertices can be manually assigned to groups. Whether groups are automatically or manually created, they can be visualized in NodeXL with unique vertex color and shape combinations that indicate membership in different clusters. Users can visualize groups in many ways including the group in a box technique [6], collapsing each group into a single unit, using summary network motifs [7], or creating subgraph images for each vertex. Combining the strategies discussed so far in Part II can facilitate the creation of insightful and visually appealing graphs that can be the basis for understanding, explanation, decision making, and persuasion.

7.6  Researcher’s agenda Strategies for dynamic filtering of complex network graphs have been around for decades, but still allow room for improvement [8–10]. Applying filters in an orderly process that arrives at a successful outcome requires skill and creative problem solving. As network researchers gain experience, they may develop more systematic approaches to choosing and setting filters so as to emphasize important features and remove distractions [11]. Being able to save a sequence of actions and then replay it on a fresh set of data would be a useful improvement. Developing standard process models (sequences of actions) to ensure complete exploration

could dramatically advance the state of the art for social media network analysis. Such a process model would be systematic yet flexible, smoothly integrating statistics and visualization [12, 13] and guiding users effectively while enabling them to explore interesting possibilities. Rapid progress in the past decade has turned the esoteric topic of clustering algorithms into a hot research area [14]. Newman’s work (see Additional resources) produced substantially improved strategies for organizing and presenting complex networks in meaningful ways, which stimulated further work on weighted, directed [15], and multiplex networks [16]. Because most algorithms run slowly, the research community has sought algorithms that can be adapted to run on multicore and specialized graphics processors that are increasingly embedded in modern computers [17]. The next step is to compute network clusters efficiently using parallel computers and cloud computing techniques. Most techniques create clusters based on edge c­onnections, but an alternate strategy is to cluster nodes by attribute values of the nodes—for example, all people who graduated from the same university are in the same cluster [18]. This introduces the visual challenge of dealing with multiple memberships.

References [1] M.E.J.  Newman, M.  Girvan, Finding and evaluating community structure in networks, Phys. Rev. E 69 (2004) 026113. [2] J.B. Lewis, K. Poole, H. Rosenthal, A. Boche, A. Rudkin, L. Sonnet, Voteview: Congressional Roll-Call Votes Database. https://voteview.com/, 2019. [3] K. Wakita, T. Tsurumi, Finding community structure in mega-scale social networks: [extended abstract], in: Proceedings of the 16th International Conference on World Wide Web (Banff, Alberta, Canada, May 08–12, 2007). WWW ‘07. ACM, New York, 2007, pp. 1275–1276. [4] M.E.J. Newman, Detecting community structure in networks, Eur. Phys. J. B 38 (2004) 321–330. [5] I.H. Witten, E. Frank, M.A. Hall, C.J. Pal, Data Mining: Practical Machine Learning Tools and Techniques, fourth ed., Morgan Kaufmann, Cambridge, MA, 2016. [6] E.M.  Rodrigues, N.  Milic-Frayling, M.  Smith, B.  Shneiderman, D.  Hansen, Group-in-a-Box layout for multi-faceted analysis of communities, in: Privacy, Security, Risk and Trust (PASSAT) and 2011 IEEE Third International Conference on Social Computing (SocialCom), 2011 IEEE Third International Conference on IEEE, 2011, pp. 354–361. [7] C.  Dunne, B.  Shneiderman, Motif simplification: improving network visualization readability with fan, connector, and clique glyphs, in: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 2013, pp. 3247–3256. ACM. [8] Q.  Li, C.  North, Empirical comparison of dynamic query sliders and brushing histograms, Proc. IEEE Symp. Inform. Vis. 2003 (2003) 147–154. [9] L.  Tweedie, B.  Spence, D.  Williams, R.  Bhogal, The attribute explorer, in: Proceedings of the CHI ‘94 Conference Companion on Human Factors in Computing Systems, ACM Press, New York, 1994, pp. 435–436.

II.  NodeXL tutorial: Learning by doing



Additional resources

[10] K.  Wittenburg, T.  Lanning, M.  Heinrichs, M.  Stanton, Parallel Bargrams for consumer-based information exploration and choice, in: Proceedings 14th Annual ACM Symposium on User Interface Software and Technology, ACM Press, New York, 2001, pp. 51–60. [11] A.  Perer, B.  Shneiderman, Systematic yet flexible discovery: guiding domain experts through exploratory data analysis, in: Proceedings ACM 13th International Conference on Intelligent User Interfaces, New York, NY, 2008, pp. 109–118. [12] A. Perer, B. Shneiderman, Integrating statistics and visualization: case studies of gaining clarity during exploratory data analysis, in: CHI ‘08: Proceedings SIGCHI Conference on Human Factors in Computing Systems, ACM, New York, NY, 2008, pp. 265–274. [13] B.  Shneiderman, Inventing discovery tools: combining information visualization with data mining, Inf. Vis. 1 (1) (2002) 5–12. [14] S.  Fortunato, D.  Hric, Community detection in networks: a user guide, Phys. Rep. 659 (2016) 1–44. [15] F.D. Malliaros, M. Vazirgiannis, Clustering and community detection in directed networks: a survey, Phys. Rep. 533 (4) (2013) 95–142.

113

[16] C.W. Loe, H.J. Jensen, Comparison of communities detection algorithms for multiplex, Phys. A Stat. Mech. Appl. 431 (2015) 29–45. [17] V.D.  Blondel, J.-L.  Guillaume, R.  Lambiotte, E.  Lefebre, Fast unfolding of community hierarchies in large networks, arXiv, 2008. 0803.0476. Available at: http://works.bepress.com/lambiotte/4. [18] A. Aris, B. Shneiderman, Designing semantic substrates for visual network exploration, Inf. Visual. J. 6 (4) (2007) 1–20.

Additional resources P.  Bedi, C.  Sharma, Community detection in social networks, Wiley Interdiscip. Rev. Data Min. Knowl. Discov. 6 (3) (2016) 115–135. S. Fortunato, D. Hric, Community detection in networks: a user guide, Phys. Rep. 659 (2016) 1–44. M.  Girvan, M.E.J.  Newman, Community structure in social and biological networks, Proc. Natl. Acad. Sci. U. S. A. 99 (2002) 7821–7826. M. Newman, Modularity and community structure in networks, Proc. Natl. Acad. Sci. U. S. A. 103 (2006) 8577–8582.

II.  NodeXL tutorial: Learning by doing

C H A P T E R

8 Semantic networks O U T L I N E 8.1 Introduction

115

8.4 Visualizing work networks

121

8.2 Creating the Twitter Gardasil HPV word pair network 115 8.2.1 Calculate word and word pair metrics 116 8.2.2 Iteratively refine the list of skipped words 116 8.2.3 Creating a new word to word network file 117

8.5 Visualizing computing dissertation and thesis connections

123

8.6 Practitioner’s summary

123

8.7 Researcher’s agenda

123

8.3 Analyzing word networks 8.3.1 Examining vertex and edge metrics 8.3.2 Examining data by groups

References

124

Suggested reading

125

119 119 120

8.1 Introduction So far in this book, you have mostly explored social networks where social actors are the vertices and edges represent connections between those people. In semantic networks, in contrast, vertices are words or concepts that are connected by co-occurrence within a pre-­defined text (such as, a tweet, a Facebook post, a news article, etc.). Semantic networks, then, map relationships between concepts and allow us to extract meaning from text based on the co-occurrence of concepts (i.e., which concepts show up together). Later in this book, you will use NodeXL to extract user-generated content from platforms such as Twitter (Chapter  11) and YouTube (Chapter 13). These chapters will apply social network analysis to find patterns of relationships between users, or between videos. The retrieved content, however, can also be used for semantic network analysis. This chapter will walk you through the process of using NodeXL’s semantic analysis features using a Twitter sample dataset. That said, the process detailed in this chapter can be applied to any text-based dataset, such as emails or news stories.

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00008-X

8.2  Creating the Twitter Gardasil HPV word pair network In this chapter, you will create and then examine a semantic network that connects words that show up in the same tweet together. The dataset is from a Twitter Search Network (see Chapter 11) that was downloaded Aug. 17, 2018. It included words related to the HPV vaccine. Specifically, the search term was: (#gardasil OR Gardasil) OR ((#HPV OR HPV) AND (vaccination OR vaccine OR immunization)). There has been significant controversy on social media about the potential dangers of the Gardasil vaccine, despite the fact that the Center for Disease Control and numerous published papers have shown it to be safe and effective.1 The original dataset connects Twitter users to twitter users. However, you will use it to create the semantic network that connects words to words as described in this section.

1 https://www.cdc.gov/vaccinesafety/vaccines/hpv/hpv-safetyfaqs.html.

115

© 2020 Elsevier Inc. All rights reserved.

116

8.  Semantic networks

8.2.1  Calculate word and word pair metrics Download the TwitterGardasilSearch.xlsx file from the book website2 and open the file. Open the Graph Metrics dialog and check the Words and word pairs metrics briefly introduced in Chapter 6. Before calculating the metrics choose Options… and select Tweet from the drop-down menu as shown in Figure 8.1. This tells NodeXL to use words found in the Tweet column on the Edges worksheet in the analysis. Make sure you have selected the Word and word pairs metric (as shown in Figure  8.1) when you click the Options… button to open the correct options dialog window. Check the Skip words and word pairs that occur only once box as shown in Figure 8.1, then click OK. It may take a while to run, as this is a large network.

8.2.2  Iteratively refine the list of skipped words Next, you will examine the results to identify words that should be omitted from the analysis in order to add

more clarity. These are often called “stopwords” in ­content analysis. A stopwords list is used to remove words that are considered to be “noise” or low value grammatical terms that occur frequently, but impart limited meaning outside of their syntactic position. A list of common American English language terms are included in the default settings for NodeXL in the Skip these words section. This list can be edited, deleted, or replaced as needed to address special topics or other languages. All unicode languages (that use the space as a word delimiter) can be analyzed using this feature. Keep in mind that this list is generic, as it is set to be relevant across social media spaces and topics. Your goal now is to refine the list by using data from the Words and the Word Pairs worksheets. Navigate to the Words worksheet to examine its contents (see the left-hand side of Figure  8.1). While the Vertices worksheet is focused on each entity and its attributes, the Words worksheet is focused on each word used in the text associated with the network (e.g., the text in the Tweet column). It reports the count of each word, its salience, and whether the word appears on

FIGURE 8.1  Graph Metrics dialog choosing Word and word pairs and setting the Options… to be based the word pairs found in the Tweet column. The highlighted words in the Skip these words section have been added. 2 https://www.smrfoundation.org/nodexl/teaching-with-nodexl/ teaching-resources/.

II.  Nodexl tutorial: Learning by doing



8.2  Creating the Twitter Gardasil HPV word pair network

any of the three Sentiment word lists. Examine the list and note which words should be removed. Focus on the most frequently used words, since the less frequently used words will likely be filtered out in later analyses. To generate a list of words to be omitted, you may ask yourself: – Are there words related to my search term that are in so many tweets they do not add any new information? (e.g., HPV, vaccine) – Are there other words related to the source of data (e.g., Twitter) that are worth omitting? (e.g., RT) – Are there other common words or numbers that likely won't add value to the analysis? (e.g., via, 2) Add the words you have identified to the Skip these words section of the Word and Word Pair Metrics dialog as shown in Figure  8.1. Recalculate the Word and word pairs Graph Metrics and continue to the next step by examining the Word Pairs worksheet (Figure 8.2). The Word Pairs worksheet is focused on words that are used together frequently in the text associated with the network. It reports the count of each word

117

pair, its salience, and whether the first or second word appears on any of the three Sentiment word lists (see Chapter 6 for a basic introduction). You can repeat the steps taken above to identify more words that may ­frequently appear in word pairs, yet do not add value to the analysis. Add them to your list of skipped words and recalculate the graph metrics. You may want to go through this process several times. For example, for a dataset about HPV the following set of keywords and standalone characters, were omitted. It took a few rounds of filtering to narrow it down: HPV URL RT vaccine vaccines vaccination VAX gtpps dr r m e g t via 2 amp. Note that these values are already added into the TwitterGardasilSearch network Options… when you downloaded the file, so you do not need to manually type them in. While this step may feel cumbersome, it is crucial for a successful semantic network analysis. Failing to remove the right keywords, especially if they are frequently used, leads to an inflation of meaningful edges in the network, making the network more connected than it is, obscuring the true structure of the network.

8.2.3  Creating a new word to word network file

FIGURE  8.2  Word Pairs worksheet showing the most common pairs of words that show up together. This can be used as the basis for a word to word semantic network.

You can now use content from the Word Pairs worksheet to create a new NodeXL file. Navigate to the Word Pairs worksheet, copy the first two columns (without the headers), and past them into the Edges worksheet of a new NodeXL file. Make sure the new worksheet is an Undirected network, since NodeXL treats word pairs as undirected and weighted (based on the Count) networks. You will also want to capture the additional data, such as the Count and Salience columns by copying them to the Other Columns section of the edges worksheet as shown in Figure 8.3. When pasting the data, you should use the Paste Special, Values (V) option that is available by right-­clicking on the cell that you want to paste. This will paste the numerical values, not the formulas. Pasting the formulas would not work in this context and would link the different files, which is not desirable. You may want to reformat the Salience column data by using the Decrease Decimal feature in the Home ribbon. At this point, there is no data in the Vertices worksheet. However, when you click on Show Graph, each unique Vertex will show up on the Vertices worksheet. You can also add Count and Salience data to the Vertices column, though making sure you line up the data correctly can be a bit tricky (see Advanced topic: Using vlookup). Once you are finished copying over the necessary data, make sure you save the new file (e.g., TwitterGardasilWords.xlsx).

II.  Nodexl tutorial: Learning by doing

FIGURE 8.3  Creating a new Semantic network file by copying data from the Word Pairs worksheet of the original social media network file to the Edges worksheet of a new semantic network file.

A D VA N C E D T O P I C Using vlookup

When using NodeXL, there is often a need to add additional data about vertices in the Other Columns. However, in many cases, the data you are copying over may not be sorted in the same order as the Vertex column data. Or, you may be pulling data from a source that includes additional vertices not included in the spreadsheet you want to insert it into. In such cases, the vlookup formula can be used to automatically populate a column of data based on the data in another column (e.g., the Vertex column). An example can help illustrate the value. Navigate to the Vertices worksheet of your new file (e.g., TwitterGardasilWords.xlsx). Rename two new columns in the Other Columns section called Count and Salience. The goal is to populate these columns with data from the Words worksheet in the TwitterGardasilSearch.xlsx network. The problem is the vertices in the Vertex column are not sorted correctly. You could sort each of them and they may line up correctly, but a safer way to make sure you have the right data associated with the right vertex is to use the vlookup function. Enter the following formula into the Count column: =VLOOKUP([@Vertex],[TwitterGardasilSearch. xlsx]Words!A$7:C$4251,2,FALSE). This formula includes several parameters. The first one, [@Vertex], says to use the data in the Vertex column as the value to lookup. For example, in the first row, it will use the word college as the index. The second parameter, [TwitterGardasilSearch. xlsx]Words!$A$7:$C$4251, proves the destination

of the lookup table. In this case it is in another file called TwitterGardasilSearch.xlsx, and includes the cells $A$7:$C$4251 on the Words worksheet. Notice the $ symbols, which make the reference stay the same, even if you copy the formula to other cells. The third parameter, 2, indicates which column in the lookup table you want to return data from. This is 2 in the Count column, but 3 for the Salience column (see formula bar in Figure  8.4), since that is the order of the columns in the original table. The final parameter is set to FALSE, which means that you need an exact match between the text in the Vertex column and the first row of the lookup table. Once the data is populated, copy the data in the two columns and use the Paste Special and Values (V) option (available when you right-click) to paste the values right over the original formulas. This will remove the formulas which included pointers to a separate file. Then, to make sure it worked properly, sort by the Count column and scroll to the bottom to see if there are any #N/A errors, which occur when no match was made. In this case, there are seven errors for words that all began with a symbol, which is throwing off the lookup feature. You can ignore the errors (assuming you don’t try and calculate anything based on them), decide to Skip these vertices if they are not important, or find the correct values in your original dataset (using the built-in Find feature of Excel) and manually copy and paste them over.



8.3  Analyzing word networks

119

FIGURE 8.4  Populating the Count and Salience columns using vlookup formula as shown in the formula bar. Right-clicking on cell AD3 and choosing Format Cells opened the dialog, where the format was changed from Text to General. This takes effect after clicking in the formula bar and typing the Enter key.

8.3  Analyzing word networks Word networks allow you to explore large datasets based on words, themes, or concepts that tend to appear together. Looking for network groups (i.e., clusters) and calculating metrics is a useful way to begin the analysis, as well as generate data that will be useful for visualizing the dataset. Begin by running the Group by Cluster feature using the Clauset-Newman-Moore algorithm as described in Chapter  7. Then calculate the following Graph Metrics as described in Chapter 6: Overall graph metrics, Vertex degree, Vertex betweenness and closeness centrality, Vertex eigenvector centrality, and Group metrics. This captures the most important undirected network metrics for analysis. A look at the Overall Metrics worksheet shows that the data is a bit messy, with some duplicate edges (which theoretically shouldn’t exist, but

may due to different textual anomalies). Change the Layout Options… so that each group is in its own box as shown in Figure 8.5 (see Chapter 7).

8.3.1  Examining vertex and edge metrics In order to understand the metrics (see Chapter  6), recall that each vertex is a word. An edge represents co-occurrence of two words within social media content (e.g., a tweet). Consider the meaning of Degree in this type of network. Since a vertex is a word, the vertex’s degree centrality measures the number of words that it appears alongside (i.e., in the same tweet) within the network. Words with high degree centrality appear with many other words, indicating the dominance of a word or concept in the overall conversation. Navigate to the Vertices worksheet and Sort Largest to Smallest on the Degree column. What are the most common words? Some are ones you would expect (e.g., gardasil,

II.  Nodexl tutorial: Learning by doing

120

8.  Semantic networks

FIGURE 8.5  Twitter Gardasil Word network with groups calculated and displayed in a different box. This image is not yet refined.

c­ ancer, girls), some are words you may not have guessed (e.g., itsmepanda1, which is a username of a prominent individual in the network), and others are artifacts of working with language that you may want to filter out by setting their Visibility to Skip (e.g., amp, which is part of a text string that indicates the & symbol). Try sorting on Betweenness Centrality. In this case, words with high betweenness centrality connect words that otherwise, are much less connected in the network. In practice, words with high betweenness centrality appear across themes of conversations. Once in a while, a word may not make sense to you. The best approach is to search for that word in the original dataset (e.g., TwitterGardasilSearch.xlsx) and find the tweets it appeared in, which will give you context as to how the word is used. A similar sorting approach can be used to identify the most important edges in the network, which represent the most important word pairs. Navigate to the Edges worksheet and Sort from Largest to Smallest on the Count column. Pairs of words that nearly always appear together, such as college pediatrician or side effects show up. While some of these may seem obvious, sometimes there are pairs that can lead you to issues that you may want to examine closer, such as forcibly injected or danish girl. Again, reviewing the tweets in the original dataset will provide context for these word pairs.

8.3.2  Examining data by groups In word networks, clusters capture sub-groups of words that appear together in social media messages (e.g., posts, tweets) more than they appear with other words. Network clusters therefore capture themes in the conversations. Navigate to the Groups worksheet to examine the various sub-conversations. The Vertices column on the Groups worksheet shows you the number of words within the group; a measure of the total vocabulary used by the group. Meanwhile, the Total Edges column shows you the number of co-occurrences of the pairs of words that show up within the group. Getting a sense of the size is important, but digging into the actual themes is most useful. To do so, focus in on the most commonly used words in each group. Navigate to the Vertices worksheet and Sort Largest to Smallest on the Count column. This will help identify the most common words. Notice that a new column now exists called Vertex Group (see Figure 8.6). The filter option can be used on this column to only display items from a particular group (e.g., Vertex Group 3 in Figure 8.6). While you don’t want to leave this type of filter on permanently, it can be a nice way of quickly focusing in on data for exploratory analysis. For example, browsing through the most common words in this cluster helps identify a theme that is focused on teen health (e.g., it includes words like girls, boys, teen, age, health).

II.  Nodexl tutorial: Learning by doing



121

8.4  Visualizing work networks

FIGURE 8.6  Filtering out all vertices that are not part of Vertex Group 3.

If the groups are too large, you may consider trying out other clustering algorithms. A companion technique to explore the dataset is to use a similar process to examine the most popular word combinations on the Edges worksheet. There you will find new Vertex1 Group and Vertex2 Group columns. Sort from largest to smallest on the Count column as in the prior example. Then filter both of those columns on the same group (e.g., 3) to show the withingroup pairs that are most common. Or, just filter on one of them (e.g., Vertex1 Group) and see what other groups the words in it connect to. In this example, you may notice there are discussions in group 3 about fda approval, 6th grade, free teens (presumably free vaccines for teens), etc. You can use a similar technique, but sort based on graph metrics. For example, the highest degree words in a group indicate the words that show up with a high variety of other words and thus, have widespread importance in the network. Those with high betweenness centrality are words that may span across groups or words within a group that serve as connecting words.

Once you have identified themes for the groups, you may want to come up with a name for each group and enter it into the Labels column on the Groups worksheet. That will allow you to visualize labels on the graph, as well as remember the work you have done.

8.4  Visualizing work networks Clear visualization is important for communicating your findings. There is no single visualization that will capture all of the insights into a dataset. Instead, you should strive to create visualizations that tell an accurate and clear picture about the words in a network. Using techniques explained earlier in Part II of the book, you can create visualizations such as Figure 8.7 to illustrate some of the most commonly used words as they appear in different sub-group conversations. When visualizing word networks, you will typically want to show the actual words as labels. As described in Chapter  5, you can set the Vertex Shape to Label. If you have calculated groups, then you will need to

II.  Nodexl tutorial: Learning by doing

122

8.  Semantic networks play

ri

pay

massdeception1 television

documentary

danish

child

story

merck

injury

grade

girls

patients

doctors

boys

cancer

thenci side

girl

outweigh

parents

outcomes

warning

proven

prevent

effects

safe

benefits

fraud

safety complaint

against

gardasil

more everyone

health

worldwide

tell

itsmepanda1

commercial

news

vaxxed

ftc

women

over

scientific

cervical

ovarian

failure

sage margin

de

very

risk

circleofdocs

doritmi

infection

wide

high

vaccin

rare

joegooding

le

parentalerte

out

damage

based

dangers

ones

rates

falling

misconduct

up

infections check fact

journal

FIGURE 8.7  Filtered view of the most common word pairs (Edge Count Greater than 4) and words (top 10 words if they are in an edge) from the 10 largest groups. Edge Width and Opacity are based on edge Count. Color is based on automatically calculated clusters. Size is based on Degree.

sonofjupit3r cplbart

vaxchoiceeast trbrad62

doritmi debnantz

real wallacehpvlab

myrightsri

potential kills

joegooding blondescientist

cns

here’s

two

hpvvaccine support

gardasil parents

takethatdoctors

okkelip

read cxloe

cancer

multiple

2008

vaxcalc kath2cats

those

FIGURE 8.8  A 1.5 degree ego-network graph for the word itsmepanda1. The word is removed from the graph. Colors are based on original groups, Size is based on Degree, and edge Width and Opacity are based on Count.

II.  Nodexl tutorial: Learning by doing



123

8.7  Researcher’s agenda

change the Group Options so that the Shape data is pulled from the Vertices worksheet, not the Groups worksheet (Chapter  7). You likely will want to use Autofill Columns to change the size of the vertices and weight of the edges so they are based on metrics or Count numbers. And you will certainly need to filter out less important edges and/or vertices to focus in on the most important words and word connections (Chapter 7). When dealing with large networks like this one, you may want to use Dynamic Filters to explore the cutoff points, and then use Autofill Columns to set the Visibility to Skip for those that you want to filter out (as described in Chapter  7). For example, in Figure 8.7 only edges with a Count value of Great than 4 are shown. Additionally, only the most common 10 vertices from each of the 10 largest groups are shown (if they are in an edge). Analysis of Figure 8.7 reveals several popular incidents that occurred during the time period. For example, the light orange group in the upper-left corner discusses an incident with a danish woman. The green group in the upper-­ middle focuses on news from a clinical trial of the National Cancer Institute (i.e., thenci). Other clusters focus on side effects, other languages, the benefits of ­vaccines, etc. Another technique is to focus in on a single word and examine the ego-network (i.e., the vertices immediately connected to it). To do so, right-click on a word of interest (e.g., the username itsmepanda1) in the graph pane and choose Select Adjacent Vertices. Next choose Toggle Selection, which will reverse which vertices were selected or not. Finally, right-click on one of the selected ones and choose Edit Selected Vertex Properties and pick Skip from the Visibility drop-down menu. You should then only be able to see itsmepanda and all of the words that directly connect with it. Finally, set the Visibility for itsmepanda to Skip, since that word is connected to all others and will obscure some of the information on the graph. You should end up with something like Figure 8.8. Analysis of the graph shows that this username is mentioned alongside several other usernames (e.g., joegooding, and most of the green colored words). It also shows that this username is mentioned alongside other words related to skepticism about vaccines, such as takethatdoctors, and kills. Creating additional ego-­ networks for other important words can reveal additional insights when they are compared.

8.5  Visualizing computing dissertation and thesis connections Word networks can be particularly valuable when there is a controlled vocabulary describing something. For example, the Proquest Dissertations and Theses

GlobalTM database3 includes data on tens of thousands of theses and dissertations. Students are asked to tag their theses and dissertations with a primary keyword and secondary keyword(s) chosen from 405 pre-determine keywords. This dataset was analyzed with data from 2004 through 2014 with a focus on dissertations and theses that related to computing disciplines [1]. A co-word analysis of the keywords helps understand the nature of computing on campuses in the United States. For example, Figure 8.9 shows the strong connections between Computer Science and Information Science, as well as Computer Science and Computer Engineering; despite the fact that there is no connection between Computer Engineering and Information Science. The important bridging role of terms such as Educational Technology, Management, and Computer Science are clearly visible and heighted in the visualization by making Size based on Betweenness Centrality.

8.6  Practitioner’s summary For social media managers, semantic networks help extract the meaning that consumers give to your brand or company. Semantic network clusters capture a variety of meanings and opinions that different conversations associate with a brand. Semantic analysis can help social media managers identify points of concern, or “red flags,” that may require intervention. Campaigns often aim to set or change opinions about a brand. Semantic networks can be used to evaluate the success of such attempts, by looking at the change of shared meaning before and after a campaign. Additionally, networks of controlled vocabularies (i.e., pre-determined keywords such as those used to tag a specific object like a dissertation) can produce particularly insightful network, or co-word graphs. NodeXL allows you to create, analyze, and visualize such co-word networks using the word and word pair metrics, as well as features described earlier in Part II of the book.

8.7  Researcher’s agenda The origins of the idea of semantic networks can be traced back to the 1960s, when researchers explored mental and cognitive processes [2], and later meaning construction, cognitive models, representation of knowledge, and semantic memory [3, 4]. Researchers argue that words are hierarchically stored in our memory and the meanings of the words depend on the relationships among

3 https://www.proquest.com/libraries/academic/ dissertations-theses/.

II.  Nodexl tutorial: Learning by doing

124

8.  Semantic networks EDUCATIONAL LEADERSHIP ADULT EDUCATION

MULTIMEDIA COMMUNICATIONS ELECTRICAL ENGINEERING

INSTRUCTIONAL DESIGN

ARTIFICIAL INTELLIGENCE

EDUCATIONAL TECHNOLOGY

COMPUTER ENGINEERING

CRIMINOLOGY

SCHOOL ADMINISTRATION

HIGHER EDUCATION ADMINISTRATION

COMMUNICATION BUSINESS EDUCATION

COMPUTER SCIENCE INFORMATION SCIENCE

HIGHER EDUCATION

TECHNICAL COMMUNICATION

PUBLIC HEALTH

LIBRARY SCIENCE

WEB STUDIES ECONOMICS

BUSINESS ADMINISTRATION

MANAGEMENT SYSTEMS SCIENCE

HEALTH CARE MANAGEMENT

NURSING

ORGANIZATIONAL BEHAVIOR

PUBLIC ADMINISTRATION BIOINFORMATICS

MARKETING

HEALTH EDUCATION

ORGANIZATION THEORY PUBLIC POLICY

OCCUPATIONAL PSYCHOLOGY

FIGURE 8.9  Co-word analysis of keywords used to describe dissertations in the Proquest Theses and Dissertation database from 2009 through 2014. Size is based on Betweenness Centrality. Edge Width and Opacity are based on the number of co-occurrences of the keywords. Only edges with a co-occurrence count greater than 7 are shown. Color is based on clusters identified with the larger network.

them. When individuals talk, they do not only express themselves but also build connections with the audience through language. The use of words in conjunction with other words creates meaning [5]. Semantic analysis is also seen as an extension of traditional content analysis [6]. Semantic networks capture the structure of co-­occurring words or concepts, providing an understanding of the meanings that people create as they discuss a topic or an issue. Semantic network analysis has become even more popular as large amounts of user-generated content have become available via the Internet. Semantic networks have been studied in a wide range of areas, from politics and health to public relations and marketing (see Suggested reading). In an era where information is distributed and consumed from a wide range of sources on social media, semantic networks are increasingly able to provide deep insights into how vocabulary is used and meanings are ascribed. Semantic networks allow scholars to trace the emergent and selfformed meanings that groups of users give to an issue, event, political candidate, health-related topic, etc.

­ o-word analysis is also possible to apply to other dataC sets, such as research publications [1]. Related methods, such as co-citation analysis, examine the relationship between publications to gain insights into research community formation and changes over time. The intersection of social networks and co-word networks is a particularly active research area that is bound to bear new fruit in the coming years.

References [1] S. Kim, D. Hansen, R. Helps, Computing research in the academy: insights from theses and dissertations, Scientometrics 114 (1) (2018) 135–158. [2] A.M. Collins, M.R. Quillian, Retrieval time from semantic memory, J. Verbal Learn. Verbal Behav. 8 (2) (1969) 240–247. [3] K.M.  Carley, D.S.  Kaufer, Semantic connectivity: an approach for analyzing symbols in semantic networks, Commun. Theory 3 (3) (1993) 183–213. [4] R.E. Rice, J.A. Danowski, Is it really just like a fancy answering machine? Comparing semantic networks of different types of voice mail users, J. Business Commun. 30 (4) (1993) 369–397.

II.  Nodexl tutorial: Learning by doing



Suggested reading

[5] M.C. McGee, The “ideograph”: a link between rhetoric and ideology, Q. J. Speech 66 (1) (1980) 1–16. [6] D. Kim, S.Y. Kim, M.I. Choi, The pivotal role of AJC in the growth of communication research in Asia: a semantic network analysis, Asian J. Commun. 26 (6) (2016) 626–645.

Suggested reading H.E. Sevin, Understanding cities through city brands: city branding as a social and semantic network, Cities 38 (2014) 47–56. Y.J. Hong, D. Shin, J.H. Kim, High/low reputation companies’ dialogic communication activities and semantic networks on Facebook: a comparative study, Technol. Forecast. Soc. Change 110 (2016) 78–92.

125

M.L. Doerfel, S.L. Connaughton, Semantic networks and competition: election year winners and losers in US televised presidential debates, 1960–2004, J. Am. Soc. Inf. Sci. Technol. 60 (1) (2009) 201–218. D.  Lycarião, M.A.  dos Santos, Bridging semantic and social network analyses: the case of the hashtag #precisamosfalarsobreaborto (we need to talk about abortion) on Twitter. Information, Commun. Soc. 20 (3) (2017) 368–385. A.  Yang, S.R.  Veil, Nationalism versus animal rights: a semantic network analysis of value advocacy in corporate crisis, Int. J. Business Commun. 54 (4) (2017) 408–430. K.H.  Kwon, C.C.  Bang, M.  Egnoto, H.  Raghav Rao, Social media rumors as improvised public opinion: semantic network analyses of twitter discourses during Korean saber rattling 2013, Asian J. Commun. 26 (3) (2016) 201–222.

II.  Nodexl tutorial: Learning by doing

P A R T

I I I

Social media network analysis case studies This section (Chapters  9–14) includes chapters that analyze networks created by a particular social media platform. The chapters need not be read in order, but they do assume you are familiar with social media and social network analysis fundamentals (Part I) and NodeXL (Part II). Each chapter in this section discusses

how to extract network data, what questions can be addressed, and how to analyze and visualize the networks to reach actionable insights. Specific chapters focus on email (Chapter  9), threaded conversation (Chapter  10), Twitter (Chapter  11), Facebook (Chapter  12), YouTube (Chapter 13), and wikis (Chapter 14).

C H A P T E R

9 Email: The lifeblood of modern communication O U T L I N E 9.1 Introduction

129

9.2 History and definition of email

130

9.3 Email networks

131

9.4 What questions can be answered by analyzing email networks? 9.4.1 Personal email network questions 9.4.2 Organizational email network questions

132 133 133

9.5 Working with email data 9.5.1 Preparing email 9.5.2 Importing email networks into nodexl

133 134 135

9.6 Cleaning email data in NodeXL 9.6.1 Remove duplicate email addresses for the same individual 9.6.2 Count and merge duplicate edges

136 136 137

9.7 Analyzing personal email networks 138 9.7.1 Creating an email overview visualization 138 9.7.2 Creating an expertise network email graph 141

9.1 Introduction Email has permeated society more than any other form of social media. It is hard to remember a time when inboxes sat on desks and spam was only a processed meat. Today, an estimated 3.8 billion worldwide email users send over 200 billion emails per day.1 Unfortunately, about half of them are considered spam.1 Email is the de facto form of communication for many corporations, nonprofits, and government agencies. Email and email lists are used to keep extended families in touch, coordinate neighborhood activities, support medical patients, 1  https://www.lifewire.com/how-many-emails-are-sent-everyday-1171210.

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00009-1

9.8 Creating a living org-chart with an organizational email network 9.8.1 TechABC's organizational unit email network 9.8.2 Normalizing and filtering TechABC's data 9.8.3 Creating an overview of TechABC's communication patterns 9.8.4 Examining TechABC's research division 9.9 Historical and legal analysis of Enron email 9.9.1 Identifying key individuals using content networks

142 143 143 143 145 146 146

9.10 Practitioner's summary

147

9.11 Researcher's agenda

148

References

148

share cutting-edge research, solve technical problems, and perform a host of other activities. In April 2017, 91% of U.S. Internet users sent or received an email message, making it the most common Internet activity of all.2 Over 80% of email users check email at least once a day.3 Unlike many social media tools, email is widely used among nearly every demographic group. Furthermore, many social media sites optionally send email notifications due to its continued ubiquity. 2  https://www.statista.com/statistics/183910/internet-activitiesof-us-users/. 3  https://www.statista.com/topics/4295/e-mail-usage-in-theunited-states/.

129

© 2020 Elsevier Inc. All rights reserved.

130

9.  Email: The lifeblood of modern communication

The integration of email into everyday life makes email networks the most accessible and in many cases most accurate source of data for mapping actual social and work relationships. Analyzing one's personal email collection is a lot like looking in a mirror. Prototype systems like PostHistory demonstrated the interest people have in seeing representations of their social media and show how maps of social connections and activity can promote sustained engagement and storytelling around important events [1]. Network visualizations provide an objective representation of one's social ties, encouraging self-reflection and providing a guide to social hygiene. These maps and reports may help you realize unappreciated or forgotten relationships, or identify a past working group that could be rekindled for a current project. They can help us overcome some of our memory biases such as weighing recent events more or remembering things we've initiated more than those initiated by others. A visualization of your personal email network can be shared with other people to explain your social world. For example, new employees could benefit from a summary of important social ties and collaborative groupings related to a job role or position. Personal email collections are of increasing importance to historians, researchers, archivists, and lawyers who are engaged in the discovery and preservation of electronic records. Analyzing organizational email networks and email lists can provide a wealth of social information that can inform important decisions and support novel interventions. Organizations can identify unique social roles, individuals who span the gaps between organizational silos, internal influencers, and employees in need of creating more connections. This information can be used as one input to help inform personnel hiring and promotion, improve retention, and spread important messages through a company. Analysis of organic employee clusters, rather than formal organizational charts, can be used to inform the formation of communities of practice and organizational restructuring, and it can help integrate relationships after mergers. Expertise networks can be identified by the use of keywords to infer topics, leading to more intelligent workgroup formation and information sharing. The analysis of an internal company or public email list can help identify experts on a topic, monitor the health of the community over time, and identify potential candidates for leadership roles in the list. Because it is based on actual behavior instead of potentially biased self-­ reports [2], its validity is high. Working with email poses several ethical challenges. Although company email is far from private and unless encrypted it is far from secure, many users don't realize just how public their email is. A 2007 survey found that nearly half of the 304 U.S. companies surveyed

monitored email use.4 More than a quarter of these companies had fired workers for email misuse. A related survey from 2006 found that 24% of employers had email subpoenaed by courts and regulators. Employers must walk a fine line between controlling the risks of litigation and security breaches by employees and not coming across as Big Brother. In such an environment, they must carefully consider the risks and rewards associated with analyzing company email collections. Researchers must also be careful to receive proper approval from list owners, managers, and members; using pseudonyms ­ when advisable;5 and protecting members' privacy. For corporations and researchers, transparency is needed when articulating the goals of the analysis, the procedures for assuring confidentiality of message content, and the decisions that will be informed by the analysis. Options for employees or research subjects to opt out or prefilter email may be desirable. Although an analysis of email collections poses some risks, social network analysis can be less intrusive than many other methods for understanding social interaction. It provides an interesting midway point for those willing to share who they talk to but not what they say.

9.2  History and definition of email Electronic mail, or email, is an electronic message transmitted over a communications network, typically as a text file with optional attachments. Email is older than the Internet itself. In the 1960s, email-like messages were sent between users of the same mainframe computer. Those who accessed the same mainframe or host computer through terminals could exchange messages. In the 1960s and 1970s, many companies used this approach to allow employees to contact other employees located throughout the world in different branch offices or subsidiaries. Email became the “killer app” of ARPANET, the computer network developed by the United States Department of Defense that evolved into the Internet. In 1971, Ray Tomlinson sent the first network email, using an “@” symbol to separate the user's name and the host computer's name. By 1973, approximately three-fourths of all ARPANET data traffic was email. Over time, email became more standardized and increasingly interoperable between different computer and network systems. 4  https://www.amanet.org/training/articles/the-latest-onworkplace-monitoring-and-surveillance.aspx. 5  Changing the user names and personal identifiers included in a network dataset is a common form of anonymization. However, it is ineffective in some situations in which a version of the original graph is available for analysis. The often-unique patterns found around each vertex in a network can be used to re-identify some anonymized entities.

III.  Social media network analysis case studies



131

9.3  Email networks

Applications for interacting with email became more feature rich and usable. Email s­ervices became essentially free from an end-user perspective with popular web services such as Hotmail, Yahoo mail, and Gmail. Although email services may have access fees, each additional email sent or received does not typically impose an additional cost to the user. Most readers are familiar with email as everyday users. Some important technical characteristics make email particularly powerful: • Flexible form. Email can be a simple plain-text message, a richly formatted newsletter, or even an interactive survey. With attachments, nearly any type of digital content can be sent, provided it is not too large a file. This flexibility allows email to support informal banter, semiformal memos, and formal letters. This flexibility also leads to email overload in which the same channel is used to host conversations, store files, track tasks, and manage transactions. • Asynchronous. The asynchronous nature of email allows people to send and receive messages on their own time, without interrupting others. The lack of immediate feedback in a text-only medium can lead to misunderstandings, but it can also encourage more careful and thorough contributions. The standard reverse-chronological order, where the newest messages are shown at the top of the inbox, makes the asynchronous nature of email more manageable by helping people distinguish new messages from old. • Broadcast. Emails can be sent to any number of people simultaneously. Ad hoc groups can be created on the fly by sending to multiple email addresses and using the common Reply to All feature. Listserv and other email list software tools allow thousands of users to communicate and listen in on the conversations of others. • Push technology. Email is considered a push technology; the sender decides what shows up in the receiver's inbox without any action on the receiver's end. This is great when trying to get someone's attention, but is also the reason so much unwanted email spam gets sent around. • Threaded conversation. Email messages are often organized into a threaded pattern consisting of messages, replies to messages, replies to replies, and so forth. This pattern mimics the natural turn-taking of spoken conversations, albeit with less frequent turnover. Threads also extend the structure of spoken interaction by enabling the creation of multiple parallel lines of conversation. If threaded properly, related messages are grouped together into a single linked collection of related messages within their context.

Services such as Usenet and discussion forums, also known as bulletin boards or web boards, share many of these characteristics making them a close cousin to email lists often called Listservs. We discuss these in detail in Chapter 10.

9.3  Email networks In a standard email network, vertices represent email addresses or corresponding people. Edges or ties are created when a message is sent from one email address to another. Edges are directed because messages are transferred from a sender to a receiver. These ties are weighted by the number of messages sent between two individuals. Table 9.1 shows a summary of the information found in seven messages pulled from Derek's personal email collection. These relationships are converted into an “edge list” and visually represented in Figure 9.1. Notice that the six rows in the Edges tab are a re-representation of the seven individual email messages shown in Table  9.1. The new edges are helpful for understanding sender-receiver relationships that are not as obvious when seen in the form of a standard list of email messages. All people in the To and the Cc email fields are counted as receivers when tallying up the Edge Weight. For example, two messages were sent from Derek to Ben; the first was sent only to Ben and the second was also sent to Marc. Derek's message to Marc copied in Anne, so an edge is created between Derek and Anne. The power of this representation is that tens of thousands of email messages among a group of people can be captured in just a few hundred rows. Alternate ways of handling the data are discussed next. TABLE 9.1  An email network edge list. From

To

Cc

Derek

Ben

HCIL Brownbag

Derek

Marc, Ben

Travel Plans

Derek

Marc

Marc

Derek

Re: Travel Plans

Carol

Derek

Tuesday meeting

Marc

Derek

Marc

Derek

Anna

Anna

Subject

Registration

Re: Registration Next Steps

This email network edge list contains seven messages from Derek's personal email collection that includes ten unique edges (including both To and Cc) and five vertices.

III.  Social media network analysis case studies

132

9.  Email: The lifeblood of modern communication

FIGURE 9.1  A simple email network visualized in NodeXL. Arrows point from the sender to the receiver(s). Edge thickness (i.e., Edge Weight) ranges from 2 to 4 and is based on number of messages exchanged. Edge opacity is set to 70. Vertex size (3 to 40) is based on Out-Degree or number of messages sent.

A standard email network can be aggregated to create networks that show the connections between different social groupings. For example, vertices can represent company work groups, organizational departments, local branches, or entire organizations. Edges can represent the aggregate number of messages sent between people associated with different groups (i.e., vertices). For example, a directed edge pointing from the marketing department to the development department with a weight of 100 would suggest that marketing employees sent 100 messages to development employees. The use of organization elements in host names (e.g., @umd.edu versus @cs.umd.edu) as part of email addresses can facilitate this type of analysis by identifying people from different departments, although the frequent use of web mail (e.g., @gmail.com) makes this technique problematic for studying broader populations. Alternatively, edges may represent the number of unique individuals who have sent emails from one department to another. For example, in our prior scenario there may have only been five people that sent the 100 messages, resulting in an Edge Weight of 5. A graph based on these networks provides an overview of the departmental relationships within an organization, highlighting the most connected departments and the most socially isolated ones. Section  9.8 provides an example of a summarized network showing connections between workgroups within a large technology company.

9.4  What questions can be answered by analyzing email networks? Email messages can be analyzed as part of a larger corpus. Table  9.2 shows the three main types of email collections (personal, organizational, and community), each of which may be analyzed by a current participant or an outside observer.6 Personal email collections include messages sent or received by an individual. Organizational email collections include messages sent and received by members of an organization. More generally, they are the aggregate of several individuals' personal email collections. Community email collections include messages sent to an email list address that get forwarded to a group of subscribed members. Email lists may be public, where anyone can participate and view prior messages, semi-public, where anyone who registers can participate and see the archive, or private, where only invited or approved members can participate and view prior messages. The goals and process of the analysis are different for each of the regions specified in Table  9.2. Outside observers such as lawyers, historians, and researchers analyze email collections for historical, research, or legal 6  Table 9.2 is loosely based on a similar figure provided by Perer, Shneiderman, and Oard who characterized the types of interactions people have with current and archival email collections [10].

III.  Social media network analysis case studies



TABLE 9.2  Types of analysis for email collections with different scales and observers.

Current Participant

Outside Observer

133

9.5  Working with email data

Personal

Organizational

Community

Region A:

Region B:

Region C:

Analyzing your own email

Analyzing your organization's email

Analyzing ongoing conversations in a community email list in which you participate

Region D:

Region E:

Region F:

Analyzing another person's email

Analyzing another organization's email

Analyzing a community email list archive in which you do not participate

reasons. In contrast, current participants such as managers, community administrators, list owners, and members analyze email collections to help inform decisions. Outside observers can benefit considerably from overviews that provide context before delving into specifics. In contrast, current participants typically understand the overall context and can delve into specifics quickly, although they may be biased in their perceptions. There are fewer privacy concerns when analyzing one's own email (Region A) or public community email lists (many communities in Regions C and F) than when analyzing organizational email (Regions B and E) or another person's email archive (Region D). We examine personal and organizational email collections in this chapter and discuss community collections in Chapter  10, since they are similar to other ­community-based threaded conversation tools like discussion forums. We discuss preparing, cleaning, and importing email data in this chapter.

9.4.1  Personal email network questions

• Temporal comparisons. How have relationships changed over time? How did an event (e.g., move to a new location) affect the network? What inactive groups exist with whom I may benefit from reestablishing contact? What projects or people have I neglected? • Structural patterns. Are there common social roles that occur among contacts (e.g., informant, decision maker, boundary spanner)? Are there types of subgroups that occur (e.g., cliques, fans)?

9.4.2  Organizational email network questions Several different questions can be asked about organizational email network datasets: • Individuals. Who are the important individuals within an organization? For example, who are the boundary spanners who link across organizational silos? Who are the influencers or topical experts? Who is not well connected and could benefit from more social ties? Who would be a good replacement for an individual? Who fills a unique niche? Who was in-the-know about an important decision? • Groups. How do email-based groupings differ from organizational structures? How is the “org-chart” different from the chart of the flow of email? How are groups interconnected? Which groups should be better connected? Is there a core competency that is not discussed in a particular branch or office? • Temporal comparisons. How does information flow through the organization? How do connections among individuals and subgroups evolve over time? How are social relations affected by a major event such as a merger or opening of a new office? • Structural patterns. What network properties are related to success? Can we identify up-and-coming stars or unique social roles based on their network structure? How is information on a particular topic distributed throughout the organization?

Several questions can be asked about personal email network datasets: • Individuals. Who are important individuals within the network? For example, who are boundary spanners who link across clusters of contacts? Who is contacted most often? Who are the most active discussants of a particular topic? Who are unwanted or troublesome correspondents? Who are unresponsive recipients? • Groups. What natural subgroups exist? What collaborative activities are individuals engaged in? What are the relationships between subgroups?

9.5  Working with email data From a user's perspective the components of an email message are relatively simple. The email header includes the From, To, Cc, Bcc, Date, and Subject fields. The email body includes the message content and any attachments. Despite this apparent simplicity, there can be a great deal of hidden complexity. A full treatment of email protocols and formats is beyond the scope of this book. Instead, we list a few key terms and facts that can serve as starting

III.  Social media network analysis case studies

134

9.  Email: The lifeblood of modern communication

points for those needing to learn more before accessing and analyzing email collections: • Email is transmitted through the Internet via the Simple Mail Transfer Protocol (SMTP). • Email uses the Multipurpose Internet Mail Extensions (MIME) format to allow character sets other than ASCII and non-text attachments to be included and transported via email. • Email client applications such as Microsoft Outlook or Apple Mail retrieve or cache messages from a mail server using Post Office Protocol (POP) or Internet Message Access Protocol (IMAP). Corporate email is typically retrieved through proprietary protocols specific to Microsoft Exchange Servers or competitors. • Email messages are stored in a variety of formats for different email clients. Some email clients store each message as a separate file; others save them in a database format. Some common formats include .eml (Microsoft Outlook, Mozilla Thunderbird), .emlx (Apple Mail), .msg and .pst (Microsoft Outlook or Microsoft Exchange), and .mbox (Mozilla Thunderbird, Gmail backup files, and many email list archives). Working with email poses technical challenges that often require preprocessing data to create useful results. The large potential size of email networks can be problematic and may require specialized programs to manage large data volumes. In practice, email will likely need to be filtered before analysis to reduce the dataset based on time ranges, people, and topics of interest. Another major challenge is the use of multiple email addresses for the same individual. In most cases, analysts are interested in social relationships between individuals, not the relationships between email accounts. The problem of combining different aliases (email addresses) for the same entity (person) is called “entity resolution,” “identity resolution,” “deduplication,” or “record linkage.” A range of tools provide deduplication services such as Marketo or the open source Python library Dedupe. Another set of tools extracts entities (e.g., names or places mentioned in email messages) which can be used to create networks that consider personal names or places mentioned in email texts rather than the sender and receiver of a message. Searching for tools that perform “named-entity recognition,” “entity identification,” “entity extraction” and “entity chunking” reveals tools such as the spaCy Python library, Stanford NER, and commercial APIs such as Lexalytics, TextRazor, ParallelDots, and Aylien.

9.5.1  Preparing email Most email clients do not export data in a format amenable to network analysis. Furthermore, the email you'd like to analyze may be stored in different formats and reside on different computers or web mail servers. As a result,

you may need to prepare your email before it is ready to import into network analysis tools such as NodeXL. The easiest way to transform email messages into network relationships (i.e., an edge list) is to use NodeXL's Import from Email Network feature. This feature relies on the Windows built-in indexing functionality on recent versions of Windows (e.g., Windows 10). By default, email files in certain formats will be indexed by Windows. You can view and change which filetypes are indexed and check indexing progress in the Indexing Options dialog accessible via the Control Panel. You may not have the email you want to analyze on a local or shared machine. For example, you may exclusively rely on a web mail service such as Gmail or Hotmail. Nearly all web mail services allow you to download local copies of your messages via POP or IMAP to an email client such as Thunderbird or Outlook, or create an archive of the files for backup. However, in some cases you may need to purchase backup software to export into a file that can be indexed. If you are using IMAP, make sure to download the complete email message files, not just the header information. Otherwise the Window's indexing service will not download the content of the messages and allow you to import them using NodeXL (as described later). You can typically choose not to download attachments if there are space limitations. If you use IMAP you can also restrict the download by folder. For example, you may want to only download recent messages (i.e., those sent in 2018) rather than years of data. After downloading messages, it may take Windows some time to index all of the files. If you have subscribed to an email list and retained all of the messages you want to analyze, you can place them in a folder and use IMAP to download just those messages.

A D VA N C E D T O P I C

Working with large email collections You may want to create networks based on email archives that are not in a format that Windows understands. For example, mbox and maildir are common formats found in Linux and Apple system mail clients. Maildir stores 100 text file per message in a directory hierarchy that matches the user's mail client, whereas mbox stores all messages in a single file. One strategy for dealing with this issue is to use a specialized programs like Aid4Mail that can aggregate email stored in multiple devices or formats, perform and store advanced searches, and export emails into a range of formats. For example, you can use Aid4Mail to open email list archive files (in .mbox format) and convert them to a format such as “eml” files that can be indexed by Windows. The tool can handle hundreds of thousands of emails with reasonable performance on a standard machine.

III.  Social media network analysis case studies



9.5  Working with email data

Another strategy is to create a database of the email messages that can be queried in multiple ways. This allows you to apply language processing and text mining approaches not available in the NodeXL import wizard. Some email programs like Aid4Mail will create a database for you. Alternatively, you can convert emails into XML and then use Excel's built-in XML maps feature to populate the Excel fields based on the XML database content.

9.5.2  Importing email networks into NodeXL Once Windows has indexed the email you want to analyze, you are ready to import the data directly into NodeXL. Select the From Email Network option from the Import drop-down on the NodeXL ribbon to open the importer shown in Figure 9.2. The enormous size of many email collections often requires filtering out messages before analysis. Even when email collections are of manageable size, filtering messages can hone in on a specific subset of messages ideally suited for addressing a question of interest. There are several ways of filtering: • Filter based on time. Include all messages sent and received during a specific time period. In NodeXL, the Date Range fields allow you to specify a time window in which messages must appear in order

135

to be included in results. Filtering based on time can be used to slice data into networks to facilitate comparison over time or measure the impact of an important event. In Figure 9.2, only messages sent between 1/1/2018 and 11/30/2018 are included. • Filter based on sender and receiver(s). Include only messages sent or received by certain people. These people may be part of a group (e.g., department, workgroup), share some characteristic in common (e.g., senior managers, located in Maryland), or have a certain relationship to another person (e.g., all those who have received a Bcc message from John). In NodeXL it is possible to specify email addresses to be found in the From, To, Cc, and Bcc fields. The default setting is a Boolean OR relationship, so if you include multiple addresses it will pull all messages with any of the addresses included. It is also possible in NodeXL to restrict messages to those that include (or don't include) any addresses in the Cc or Bcc fields through the checkboxes available the righthand side of Figure 9.2. Filtering based on the sender and receiver(s) is ideal for focusing in on a subgroup of important people for further analysis. • Filter based on content. Include only messages that share some characteristic or content of interest. In NodeXL messages with or without attachments, messages within a certain size range, and messages or subjects with specified text can be filtered. The

FIGURE  9.2  NodeXL Import from Email Network dialog filtered to only include messages with the term “cybersecurity” sent between 1/1/2018 and 11/30/2018.

III.  Social media network analysis case studies

136

9.  Email: The lifeblood of modern communication

text search feature can be powerful when combined with standard naming conventions. For example, all messages from the Association of Internet Researcher's email list can be selected by searching for the text string “[air-l]” in the Subject search box. In the example shown in Figure 9.2, only messages with the word “cybersecurity” in the message body are included. This was chosen because a new Cybersecurity major was started at BYU in 2018 allowing a more focused analysis of this topic. • Filter based on folders and labels. Include only messages that are found in a specified folder or messages with a certain label (e.g., in Gmail). People often organize or label (i.e., tag) email into meaningful collections based on subject matter or projects. In NodeXL, messages can be restricted to those found within a certain folder. Filtering based on folders is ideal for capturing interactions about a specific project or topic that may not be identifiable from a simple keyword. The “Sample folders” link shown in Figure 9.2 shows examples of pathnames that can be entered. If you do not know the name of your email account, you can open up Outlook, rightclick on the folder you are interested in, and choose Properties. For example, if I were to restrict my search to all messages in the Sent Items folder, I would type in /[email protected]/Sent Items. • Filter based on a combination of features. The various filtering options can be combined in intricate ways to find, for example, messages from a subset of key people sent during an important time period with a certain keyword. If more advanced filtering is needed, you can use an advanced email management tool like Aid4Mail to create a folder of desired messages and restrict the import to that folder. This enables the use of advanced search queries (e.g., regular expressions) for identifying messages (see Advanced topic: Working with large email collections). In addition to filtering the messages included in the social network dataset, NodeXL allows the way the edge weight is calculated to be specified either based only on addresses in the To field or including those in the Cc or Bcc fields as well. This is independent of filtering. By default only those addresses in the To field are counted. In the example displayed in Figure 9.2, the Cc field is included when calculating edge weights, but not the Bcc field.

9.6.1  Remove duplicate email addresses for the same individual If your focus is on connections between people, as opposed to specific email accounts, you will want to combine multiple email accounts from the same person into a single one. Unless you are using an advanced entity resolution software program to do this, this is likely to be a somewhat manual process. The simplest approach is to use the Find and Replace tool familiar to most Excel and Word users. Start by choosing Show Graph and then navigate to the Vertices worksheet. Sort the Vertex column from A to Z so that email addresses that start with the same name will be next to one another (e.g., [email protected] and [email protected]). Then click on Control+F to open the Find and Replace window and enter the appropriate email addresses (Figure  9.3 presents an example). Navigate to the Edges worksheet and choose Replace All for data in the Vertex1 and Vertex2 columns. There will likely be duplicates that don't start with the same username (e.g., dlhansen@byu. edu is my work email address, while shakmatt@gmail. com is my personal email address). To find the important people based on the frequency of interaction you can sort columns by edge weight and make sure that those with a high edge weight are not duplicates. The most important duplicates to remove are your own. After replacing the fields in the Edges worksheet, you should delete all of the rows that have data in the Vertices worksheet. Then click Show Graph, which will generate a new list of Vertices on the Vertices worksheet. If you fail to do this, you will have duplicates in the Vertices worksheet, which can cause problems later. The problem with using Find and Replace is that it must be repeated each time the data is re-imported or updated, even if the email addresses are the same. There is also no trace of the changes once they are made, making it hard to audit mistakes. A more time intensive, but careful, approach is to use a Lookup Table as described in the Advanced topic: Performing the lookup table strategy to count and merge duplicate email addresses.

9.6  Cleaning email data in NodeXL After importing email data into NodeXL, you will likely need to clean it to remove duplicate email addresses for the same individuals, as well as self-referring loops created when people reply to their own messages.

FIGURE  9.3  Excel Find and Replace dialog used to help combine different email addresses in the NodeXL Vertices and Edges worksheets.

III.  Social media network analysis case studies



9.6  Cleaning email data in NodeXL

137

A D VA N C E D T O P I C

Performing the lookup table strategy to count and merge duplicate email addresses Once you have imported email data into the Edges worksheet and chosen Show Graph, you can navigate to the Vertices worksheet where there is now a list of all of the unique email addresses. Copy that list to a new worksheet and title the column Original_Addresses. Create a new column next to it called New_Addresses. When you find duplicate addresses for the same person in the Original_Addresses column, repeat the desired address in the New_Addresses column. An example is provided in the Lookup_Addresses table shown in Figure 9.4. You can now use a VLOOKUP function to look up the New_Addresses that correspond with the Original_ Addresses to create a new edge list with no duplicate email addresses. Copy the original edge list from the Edges worksheet (columns D and E) and create two new columns for a new edge list (columns F and G). The New Edge List columns will be automatically populated by the results of a VLOOKUP function. Figure 9.4 shows an example in Cell F9. The VLOOKUP function looks for the original Vertex1 address found in cell D9 ([email protected] shown in the blue outlined rectangle) within the first column of

the Lookup Addresses table (cells $A$3:$B$7 shown in a red outlined rectangle). In this example, it finds an exact match in cell A7. It then returns the value of the second column in the Lookup table (because the number 2 was entered into the VLOOKUP formula), which is dlhansen@ byu.edu found in cell B7. The FALSE in the VLOOKUP formula specifies that the value that is looked up must exactly match a value in the first column of the Lookup_ Addresses table. An error is returned if an exact match is not found. Once the New Edge List is created, you can copy the new Vertex1 and Vertex2 columns (F and G) and use Excel's Paste Special feature to paste their values into a new workbook. You should also copy and paste the Edge Weights column from the original file after assuring that the Vertex1 and Vertex2 columns are in the exact same order as the original data. VLOOKUP functions can use considerable computational resources when working with large files, so once they are calculated you may want to copy them, choose Paste Special, and select Values so they do not need to be recalculated each time changes to the workbook are made.

FIGURE 9.4  A Lookup table and the Excel = VLOOKUP() formula is used to combine multiple email addresses associated with the same person in NodeXL.

9.6.2  Count and merge duplicate edges Once you have updated the email addresses in the Edges worksheet so that different addresses for the same person are replaced with a single address, you will likely have duplicate edges (e.g., more than one row that have the same values in the Vertex1 and Vertex2 columns). It can be useful to “roll up” these duplicate edges, replacing multiple connections between a pair of email addresses with a single edge. The rolled up edge has a weight equal to the total number of exchanged messages found in the data. It is important to roll up the data so that network metrics can be accurately calculated, as some of them assume that edges connecting any pair of

vertices are unique. To prepare your email network for analysis, you can roll up repeated email messages from the same pair of people using the Count and Merge Duplicate Edges feature in the Prepare Data section of the NodeXL Ribbon. This will merge the duplicate edges and sum up the Edge Weights so the total Edge Weight remains the same. Before doing this, make sure the network type is set to Directed, or else it will remove the directed nature of the graph. Figure  9.5 shows the New Edge List from Figure  9.4 in Columns A and B and a merged version of it in Columns E and F shown here to illustrate the results of the Count and Merge Duplicate Edges feature. Notice that the total of the Edge Weight column is the same.

III.  Social media network analysis case studies

138

9.  Email: The lifeblood of modern communication

FIGURE 9.5  Effects of NodeXL's Count and Merge Duplicate Edges feature after combining duplicate addresses. A self-loop is shown in red.

A D VA N C E D T O P I C

Automatically identifying self-loops To easily identify self-loops, you can create a new column on the Edges worksheet called Self-Loop and populate it with the function = Edges[[#ThisRow], [Vertex1]] = Edges[[#ThisRow], [Vertex2]]. If the Vertex1 and Vertex2 column are the same for a given row, a TRUE will be returned. Otherwise a FALSE will be returned. Once calculated, you can sort on the column to find the Self-Loops. If desired, you can delete them or choose to skip them using the Visibility column.

vides more detail on the workgroup that he will work with most closely. In the following examples, the new employee is a new faculty member coming to work with Derek Hansen, Associate Professor at Brigham Young University's IT and Cybersecurity program. The faculty member will be focused on the area of cybersecurity. For privacy reasons, the email networks analyzed in this section are not made public. You are encouraged to analyze your own email data with a similar scenario in mind. This section assumes that Windows has already indexed your emails as described in prior sections.

9.7.1  Creating an email overview visualization Sometimes people send email messages to themselves as a reminder, as a To Do list, or to share a file between computers. This results in a row with the same address in the Vertex1 and Vertex2 columns on the Edges worksheet and is called a self-loop. The use of multiple email addresses and the removal of duplicate addresses can also cause self-loops. Row 9 of Figure 9.4 is an example. For many analyses these self-loops are not important and can be distracting when visualizing data or calculating network metrics. You may want to remove self-loop edges such as the red pair in Figure 9.5 as an additional step after you have counted merged duplicate edges. See Advanced Topic: Automatically identifying self-loops.

9.7  Analyzing personal email networks This section presents two projects that serve as examples of how to analyze personal email collections. They are both based on the following scenario. Scenario: You have a new employee coming to work with you next week whom you will supervise. He doesn't know you well and is new to the organization. To help him smoothly transition into his new job, you want to provide him with an overview of the people you work with and their relationships to each other. You decide to create two network visualizations, one that provides an overview of all of your contacts and another that pro-

Step 1: Import data into NodeXL Import all of your email sent within the past month. Although some people you know may not have contacted you in the prior month, this time period lets you collect a broad set of your active email contacts. Use the Import From Email Network feature in the Data menu and filter based on your chosen date range (e.g., 11/1/2018 to 11/30/2018). Check the Use Cc line when calculating edge weights box to be more inclusive. For Derek's dataset, a total of 1977 edges are created with a total edge weight of 7572. Step 2: Clean data Next, combine email addresses as described in the Advanced topic: Performing the lookup table strategy to merge duplicate email addresses, and run the Count and Merge Duplicate Edges function. For my dataset, this collapsed the 1977 edges into 1837 (140 pairs were merged). To make sure no data was lost, check that the sum of the Edge Weight column is the same as it was before the merge. Step 3: Filter data To more clearly focus on the key relationships, it is desirable to remove infrequent email exchanges. Sort the Edge Weight column from largest to smallest. It is likely that the values will have a skewed ­distribution

III.  Social media network analysis case studies



9.7  Analyzing personal email networks

with many connections with very low edge weights and relatively few connections with a high edge weight. Remove the least common connections by choosing a cutoff point for deletion. You may want to use the Dynamic Filters feature discussed in Chapter 7 to find an appropriate cutoff. For example, in Derek's data when all of the connections with an edge weight of 100,000 employees in dozens of countries and hundreds of locations. Employees are aggregated into roughly 10,000 organizational units that have an average of 15 members. Organizational names in the visualizations have been anonymized. For privacy reasons we cannot provide the dataset. People in each organizational unit send and receive email from people within their own unit as well as to people in other units. These events were logged in the corporate email server and were extracted for a weeklong period. An edge list of events in which an employee sent an email to another employee in the To, Cc, or Bcc fields was created. Data about each employee were then removed and replaced with the name of the organizational unit in which they were a member, helping address individual privacy concerns. Data were then aggregated (see Section  9.6.2), creating an edge weight that represents the number of messages sent from one unit to another. This process rolls messages exchanged between members of the same unit into selfloops, where the sending unit and receiving unit are the same. The total number of internally exchanged messages can be useful, but is best treated as attribute data on the Vertices worksheet rather than captured in the edge list.

9.8.2  Normalizing and filtering TechABC's data Whole graph maps of enterprise networks are likely to be too large and dense to be informative. For example, TechABC's raw sent email network includes >1.3 million edges and around 10,000 vertices. A process of filtering and selective display is required to peel away parts of the network that obscure structures of interest (see Chapter  7). When working with large datasets such as TechABC's, you may want to perform the first round of filtering using a database program like Microsoft Access because of size limitations in Excel. A common edge filtering technique is to remove all connections below a threshold, helping whittle away infrequent ties to reveal the strong core skeletal structures of the company. The easiest threshold to use is the raw number of messages sent between units. However, because organizational units differ in size, this approach disadvantages smaller units with fewer members contributing to the number of messages. To account for this discrepancy, you can normalize the data by creating a new edge variable based on the number of messages sent per employee (e.g., per full-time equivalent or FTE). You'll need to decide

143

if you want to use the number of FTEs from the sending unit, receiving unit, or some combination of the two. For the graph shown in Figure  9.8, we removed edges with fewer than 50 messages per FTE sent in a week, where we used the minimum of the sender and receiver FTE values as the denominator. This approach keeps an edge if it is important (i.e., a high number of emails per FTE) to either the sending or receiving unit (see the U.S. Senate co-voting example in Chapter 7 for another illustration of a similar technique). The resulting, filtered TechABC network includes 2303 edges and 2267 vertices. Figure 9.9 uses a similar approach, but because it focuses on a subset of units (only research units), the threshold was lowered to 10 messages per FTE sent in a week. You could also normalize the data by calculating the number of messages sent from one unit to another unit as a percentage of all messages sent from the unit. This approach accounts for differences in a unit's overall email usage patterns, which can be desirable in some cases. For example, it would remove edges representing company announcements from a single unit (e.g., the human resources or information technology department) because the messages sent to any one unit would be a small percentage of the sending unit's overall sent messages. As with the prior example, you will need to decide if you want to use the sending or receiving unit's total message count as the denominator, or some combination of the two (e.g., maximum, minimum, average). Other strategies for filtering data can lead to other insights. For example, showing only weak ties (edges with between 3 and 10 messages per FTE) can highlight ­lesser-known connections that might guide management efforts to improve connections across gaps in the company. Attributes of organizational units can be used to filter the network graph as well, helping to zoom into subsections of the larger graph. For example, you could remove all but the most central groups to reveal the network of core groups while hiding more peripheral groups. You can also focus on units within a particular department, geographic location, or similar mission. You will see this approach used in our second example (Figure 9.9), which looks at connections between research units of TechABC.

9.8.3  Creating an overview of TechABC's communication patterns You may want to create an overview graph of an organization's email communication before moving into more detailed analyses of specific departments or groups. Overview graphs can be difficult to read because of their size. However, they are excellent for dynamically exploring by sorting on metric properties to identify important units, highlighting vertices of interest and seeing their connections on the graph, and using dynamic filters

III.  Social media network analysis case studies

144

9.  Email: The lifeblood of modern communication

FIGURE 9.8  TechABC's organizational unit email network “backbone,” focusing on high-traffic connections between units (i.e., those who

exchange >50 messages per FTE). Color is mapped to betweenness centrality with green vertices playing important roles as bridge spanners. Edge opacity is mapped to messages per FTE. Dynamic filters were used to exclude those with low closeness centrality, which is a trick for filtering out all vertices that are not part of the large component.

to further hone in on specific areas of interest. A highly filtered overview of TechABC is shown in Figure  9.8. Only edges with >50 messages per FTE are shown, with additional filtering to show only the main component. You can think of this as the backbone of the company. This graph and the accompanying data tell interesting stories about the company. Overall, the graph is sparse, largely because of the high filtering threshold we have chosen. The graph density is very low, suggesting that most units only communicate heavily with one or two other units. The average geodesic distance is 10.2 and maximum geodesic distance (i.e., diameter) is 29, both of which are quite high. If high numbers existed at a lower threshold, it would suggest that units may not be well connected with certain other units on the other “side” of the company. Increasing connections between otherwise disconnected groups may be a goal for an organization. For example, many organizations have created “communities of practice” consisting of people with similar skills who

are scattered throughout different organizational units. An initiative to increase connections throughout the company could be evaluated by looking for increases in the network density and decreases in the diameter over time. In addition to looking at global trends, it is possible to look at the role of individual units in the company. Even with the highly filtered graph shown in Figure 9.8, it is possible to identify several hubs (with high out-­degree), some densely connected clusters, and units that act as bridges between other units. Some of these fill critical locations in the network, demonstrating their unique value that results from their network position. Organizational units that are connected to many other units (i.e., the hubs) perform services like IT management or library services that touch many parts of the company. Groups that are less connected but have a high betweenness centrality likely have coordination functions, bridging information between multiple groups such as specific geographical units within a larger ­region. Isolated

III.  Social media network analysis case studies



9.8  Creating a living org-chart with an organizational email network

groups and clusters of groups are likely specialists that perform a function for one or a few other groups to consume. Analyzing networks like the one visualized in Figure  9.8 also allows you to compare units that serve a similar function to see how they compare on various metrics, helping to identify those that could benefit from additional connections.

9.8.4  Examining TechABC's research division Although overview maps like Figure 9.8 can be helpful, they can also be cluttered and may filter out too much of the detail for large companies. To gain actionable insights, you will typically need to focus on subsections of the network, such as units that serve a similar purpose (e.g., IT, marketing, research). In this section you will explore the organizational units within TechABC that have a research mission. They were identified by looking for organizational unit names with the word “research”

145

in them using Microsoft Excel's Search function (a non-case-sensitive function similar to the Find function). Although you can restrict the network analysis to only the core units that meet your criteria (i.e., research units), it is often insightful to include all units connected to the core units. For example, Figure  9.9 includes all research units (maroon boxes), as well as all units they sent or received messages to (blue disks). A cutoff point of 10 messages per FTE was used in order to account for unit size differences. Because the focus is on the research units, connections between the non-research units are not shown. The result is a collection of all of the 1.0-degree networks of the research units. To create a similar graph, we added a new column called Research to the Edges worksheet that is a 1 if either Vertex1 or Vertex2 is a research unit and a 0 otherwise. This can be set to Edge Visibility Equal to 1 (in the Research column) using the Autofill Columns feature to exclude all other edges.

FIGURE 9.9  TechABC's organizational unit network including research units (maroon squares) and non-research units connected to them (blue disks) through the exchange of email. Only edges with 10 or more messages sent per FTE are included. Edge width (1–2) and opacity (40–100) are based on raw sent messages. Vertex size is based on the number of group members (FTEs). Notice some obvious disconnects (e.g., Market Research 1 and 2), as well as some of the important bridge spanning research groups (e.g., Specific 6) and non-research groups (e.g., Blue disk just above General 17).

III.  Social media network analysis case studies

146

9.  Email: The lifeblood of modern communication

The network highlights several important bridge-spanning units, as well as some disconnected units that may need to be connected. For example, research unit Specific 6 plays an important role in connecting several other research groups either directly or indirectly. The organizational unit Specific 10 is important because it is the only path connecting the large Specific 2 unit to the other research units (albeit indirectly). There are also several non-research groups that play pivotal bridge spanning roles, such as the very small unit just above General 17 that is connected to six different research groups, none of which is directly connected to another. This small unit likely plays an important role and its small size may make it vulnerable to employee turnover, suggesting that the company may consider if additional resources are needed to support the group's function. In contrast, the network shows Market Research 1 and 2 in completely different components, not even connected indirectly. More generally, few research units are directly connected to each other, suggesting that there may be potential for increased exchanges through employee swaps, internships, or other shared projects. This assumes there would be benefits from interdisciplinary projects, which content experts would need to determine. Although many of the actionable insights require knowledge about the organization, Figure 9.9 gives you some idea of the potential benefits of this type of analysis.

9.9  Historical and legal analysis of Enron email In the prior section you explored an organizational network from the perspective of an insider who knows the company. In this section you will explore an organizational network from the perspective of an outsider trying to make sense of an email corpus collected as part of a lawsuit. Specifically, you will explore a subset of email messages sent and received by Enron employees. The original, publicly available dataset included approximately a half-million messages and was made public by the Federal Energy Regulatory Commission (FERC) during the investigation of Enron. It was later cleaned and made permanently accessible by researchers at MIT, CMU, and SRI (see www-2.cs.cmu.edu/~enron for details). The analysis in this chapter is based on a subset of 1700 messages coded by students and researchers at the University of California at Berkeley, filtered to only include messages that are work related. It focuses on ­business-related messages occurring later in the collection and includes discussions of the California Energy Crisis (see http://bailando.sims.berkeley.edu/enron_email. html for a complete description and compressed file of the individual messages). Messages were downloaded,

indexed, and imported into NodeXL using the process described earlier in this chapter. You can download the NodeXL files that correspond to the images shown in this section from https://www.smrfoundation.org/nodexl/ teaching-with-nodexl/teaching-resources/. The analysis is inspired by Jeffrey Heer's work [4].

9.9.1  Identifying key individuals using content networks One problem historians and lawyers face is identifying individuals who played key roles in important events. For employees that use email frequently, email networks provide a quick sense of who communicates with whom. Filtering email collections to include only those that use a particular keyword or set of words is a useful method for finding people related to some event. You can see an example of this by analyzing the Enron email network of messages that include the term “FERC,” the commonly used acronym for the Federal Energy Regulatory Commission, an “independent agency that regulates the interstate transmission of natural gas, oil, and electricity” (see www.ferc.gov). To create this FERC network, you can use the NodeXL import tools, making sure to filter messages to include those that have “FERC” in the body of the message. The resulting network includes 370 vertices representing employee email addresses and 672 weighted edges. This is a smaller subset of the Enron message network tagged by UC Berkeley students that includes 1803 edges and 1102 vertices. The total sent and received FERC messages8 are included on the Vertices worksheet (see Advanced topic: Calculating total sent and received edges), along with a column called %_Received, which equals Received/ (Sent + Received). Once you calculate the graph metrics, you can use them to create a graph such as Figure 9.10 designed to highlight important individuals. The graph sets the size of each vertex based on in-degree, because those receiving FERC messages from many different individuals are likely “go to” people. Vertex color is based on the %_Received data, with greener vertices indicating that the individual received many messages but did not send out many. Individuals with something to hide may not send out messages, suggesting that focusing on the large green vertices may lead to potential violators. Indeed, one of these vertices represents Tim Belden, the head of trading in Enron Energy Services considered by many to be the mastermind of Enron's scheme to drive up energy prices in California. Belden pleaded guilty to one count of conspiracy to commit wire fraud as part of a plea bargain and ended up serving as a key witness against many top Enron executives. Although visualizations like Figure  9.10 can help identify individuals worth following up on, they should

III.  Social media network analysis case studies



147

9.10  Practitioner's summary

FIGURE 9.10  Enron Corporation's network of email including messages with the word “FERC” exchanged between employees. Vertex size is based on in-degree. Greener vertices received many FERC messages but did not send many out. The Harel-Karen layout was initially used, followed by Fruchterman-Reingold, to push the more peripheral vertices to the edges. Tim Belden, who pleaded guilty and witnessed against other top Enron executives, is labeled.

be used cautiously. In this particular example, there are no messages sent from Tim Belden in the dataset, making it unclear if his high received ratio is due to his actual email usage patterns, purposefully deleted messages, or limitations with the original dataset. Even if the data accurately reflects actual email patterns, Figure  9.10 is imperfect in that it emphasizes many individuals aside from Tim Belden who were not accused of illegal activities. Furthermore, many of those found guilty of crimes were not included in this graph at all, perhaps because they recognized the liability of using email for sensitive communication or perhaps because of limitations in the dataset. Clearly, reading the content of the messages is of utmost importance. However, viewing the network can help identify individuals and messages of interest. Once an individual is known to be involved, mining email is an effective way to identify people with whom the suspect frequently interacts. For example, Figure 9.10 shows a strong connection from John Shelk to Tim Belden (and many other recipients), which is explained by the fact that John Shelk often reported on congressional meetings but rarely received replies to his reports. Integrating

the content with network visualization tools can provide a powerful exploratory platform, as has been done with the Enron network dataset [4].

9.10  Practitioner's summary Email networks provide an intimate look into individuals' social and work relationships making them of interest to managers, community analysts, historians, researchers, and legal professionals. Because email is frequently and widely used in professional contexts, it reliably captures important aspects of many professional relationships. There are three main types of email collections: personal, organizational, and community. An analyst's existing experience with a collection is also important and impacts the types of questions asked and amount of detail needed. Working with email networks can be challenging. Large collections must often be filtered to a manageable size. Filtering can be based on time, sender/receiver, messages' content, folders or labels, or any c­ ombination.

III.  Social media network analysis case studies

148

9.  Email: The lifeblood of modern communication

Combining duplicate email addresses for the same individual can be time intensive but is often necessary. Integrating email networks with corporate personnel data can be challenging and poses ethical considerations, but when done responsibly can provide new insights. Personal and organizational email networks can be explored for insights or shared with others to provide an overview. These networks may be based on individuals and their connections or on organizational units and their connections. Analysis can uncover important individuals and relationships such as boundary spanners, central members, broadcasters, and unresponsive recipients. Tightly connected subgroups can be identified and their relationship to one another can be mapped. The impact of interventions or external shocks on the network can be tracked over time, and common structural patterns such as recurring social roles or types of subgroups (e.g., cliques, fans) can be identified. These analyses can lead to actionable insights by identifying people or departments that need more cross-fertilization, helping newcomers get an overview of the social structure they are entering into, evaluating the effectiveness of a new community of practice initiative, and much more.

9.11  Researcher's agenda The widespread use of email has fostered a growing community of researchers whose goals are to understand usage patterns so as to improve user interfaces and management tools. Researchers have focused largely on individual usage of email [5, 6], but they increasingly work on forensic tools to analyze other person's email or a group's email [3, 7]. A popular theme has been to improve the strategies for finding relevant documents in a large email collection [8, 9]. Exploration tools have built on the traditional keyword or key phrase search strategies, but increased attention to visualization tools has enabled users to get an overview of temporal patterns, relationships with individuals, or the social structure within groups [8, 10–13]. The many opportunities to improve on email analysis systems is generating increased research on these issues and an increasing demand for such tools from corporate human resources staff, forensic investigators, legal analysts, and social scientists. The ability to detect temporal changes, such as sharp increases/decreases in

c­ ommunication among certain people or about certain topics, is a valuable guide to analysts. Temporal changes might be visualized by simple timelines or by animated changes to network diagrams, assuming stable layouts are used. The formation and dissolution of subgroups signal important changes that are useful in applications as diverse as detecting rumor spreading (gossip), corporate reorganizations, or antecedents of important events. Tying email to geographical position or even location in office buildings could help us to understand social processes in organizations.

References [1] F.B.  Viégas, D.  Boyd, D.H.  Nguyen, J.  Potter, J.  Donath, Digital artifacts for remembering and storytelling: postHistory and social network fragments, in: Proceedings of Hawaii International Conference on System Sciences (HICCSS), 2004, pp. 105–111. [2] S.I.  Donaldson, E.J.  Grant-Vallone, Understanding self-report bias in organizational behavior research, J. Bus. Psychol. 17 (2002) 245–260. [3] A. Leuski, Email is a stage: Discovering people roles from email archives, in: Proc SIGIR 2004, ACM Press, New York, 2004, pp. 502–503. [4]. J. Heer, Exploring Enron: Visualizing ANLP Results, Available online at: http://hci.stanford.edu/jheer/projects/enron/v1. [5] N.  Ducheneaut, V.  Bellotti, Email as habitat: An exploration of embedded personal information management, Interactions 8 (5) (2001) 30–38. [6] S.  Whittaker, Personal information management: from information consumption to curation, Ann. Rev. Inform. Sci. Technol. (2013) 1–62. [7] J.R.  Tyler, D.M.  Wilkinson, B.A.  Huberman, E-mail as spectroscopy: Automated discovery of community structure within organizations, Informat. Soc.: Int. J. 21 (2005) 143. [8] G. Tang, J. Pei, W.S. Luk, Email mining: tasks, common techniques, and tools, Knowl. Inform. Syst. 41 (1) (2014) 1–31. [9] D.  Elsweiler, M.  Baillie, I.  Ruthven, Exploring memory in email refinding, ACM Trans. Information Systems 26 (4) (2008) 1–36. [10] A. Perer, B. Shneiderman, D.W. Oard, Using rhythms of relationships to understand e-mail archives, J. Am. Soc. Inf. Sci. Technol. 57 (14) (2006) 1936–1948. [11] A.  Perer, M.A.  Smith, Contrasting portraits of email practices: Visual approaches to reflection and analysis, in: Proceedings International Conference on Advanced Visual Interfaces (AVI 2006), 2006, pp. 389–395. [12] S.J. Luo, L.T. Huang, B.Y. Chen, H.W. Shen, Emailmap: visualizing event evolution and contact interaction within email archives, in: 2014 IEEE Pacific Visualization Symposium, IEEE, 2014, pp. 320–324. [13] F.B.  Viegas, S.  Golder, J.  Donath, Visualizing email content: Portraying relationships from conversational histories, in: Proceedings CHI 2006, ACM Press, New York, 2006, pp. 979–988.

III.  Social media network analysis case studies

C H A P T E R

10 Thread networks: Mapping message boards and email lists O U T L I N E 10.1 Introduction

149

10.6 Understanding groups at Ravelry

158

10.2 Definition and history of threaded conversation 149

10.7 Practitioner's summary

159

10.3 What questions can be asked

151

10.8 Researcher's agenda

159

10.4 Threaded conversation networks

152

References

160

10.5 Identifying important people and social roles in the CSS-D Q&A reply network

Further reading

160

154

10.1 Introduction Threaded conversations have served as the foundation of virtual communities since the inception of the Internet. Usenet newsgroups, email lists, web boards, and discussion forums demonstrated the value of threaded conversation from the beginning. More ­recent incarnations of threaded conversations show up in Facebook and LinkedIn group and profile page discussions, Reddit threads, GameSpot, Craigslist posts, YouTube comments, Amazon ratings, and Q&A sites like Quora and Stack Overflow. All contain collections of messages sent in reply to one another. The natural conversation style supported by the basic post-and-reply threaded message structure has proven enormously versatile, serving communities ranging widely in focus and goals. Cancer survivors and those seeking technical support or religious guidance are as likely to use a threaded discussion as a corporate workgroup. Although the basic structure of threaded conversation has remained surprisingly similar over time, conversations now include multimedia elements, user profiles (often with social network features), participation statistics, reputations scores, and ratings. Conversations are now attached to a host of other entities ranging from people (e.g., a public wall on a person's profile page) to items (e.g., ­movies;

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00010-8

actors) to groups (e.g., university alumni) to events. This chapter primarily focuses on more traditional forums and email lists, but the core analysis techniques for threaded networks apply to other contexts. The threaded conversation structure lends itself well to network analysis, because a directed link between individuals is created each time someone replies to another person's message. Unfortunately, most threaded conversation systems do not make this networked data easily accessible. The majority of threaded message content is not easily accessible because of the number of software platforms used and the fact that many groups only make content accessible to subscribed members. Many threaded message systems report participation statistics and ratings (e.g., top 10 contributors), which are important metrics. However, they fail to capture the social connections between members, a critical component of virtual communities and internal communities of practice.

10.2  Definition and history of threaded conversation Threaded conversation is a commonly used design theme that enables online discussion between multiple participants using the ubiquitous post-reply-reply

149

© 2020 Elsevier Inc. All rights reserved.

150

10.  Thread networks: Mapping message boards and email lists

s­ tructure. The key properties of threaded conversation were enumerated in Resnick et al. [1] and are listed here with some modification: • Topics: A set of topics, groups, or spaces, sometimes hierarchically organized to aid users in discovering interesting groups to “join.” Topics or groups are persistent, though their contents may change over time. Figure 10.1 includes two topics: TOPIC 1: Social Media and TOPIC 2: NodeXL. • Threads: Within each topic or group, there are toplevel messages and responses to those messages. Sometimes further nesting (responses to responses) is permitted. The top-level message and the entire tree of responses to it is called a thread. In Figure 10.1, there are five unique threads (starting with a green background box). Thread A includes only two messages, whereas Thread B includes six messages. Thread D includes only a single message. • Single authored: Each message contributed to a thread is authored by a single user. Typically, the person's username, real name, or email address is shown alongside the post so people know who is talking.

TOPIC 1: Social Media Thread A | Adam | 12/10/2010 2:30pm Re: Thread A | Beth | 12/10/2010 5:30pm Thread B | Cathy | 12/13/2010 11:00am Re: Thread B | Dave | 12/14/2010 3:30am Re: Re: Thread B | Beth | 12/14/2010 12:00pm Re: Thread B | Ethan | 12/15/2010 10:00am Re: Re: Thread B | Dave | 12/15/2010 1:30pm Re: Re: Re: Thread B | Cathy | 12/16/2010 1:00pm

TOPIC 2: NodeXL Thread C | Dave | 12/10/2010 9:30am Re: Thread C | Beth | 12/11/2010 10:00am Thread E | Fiona | 12/14/2010 9:00am Re: Thread E | Fiona | 12/14/2010 9:05am Thread D | Greg | 12/13/2010 8:00am

FIGURE 10.1  Threaded conversation diagram showing five threads that are part of two different topics. Each post includes a ­ subject (e.g., Thread A), a single author (e.g., Adam), and a time stamp (e.g., 12/10/2010 2:30 pm). Indenting indicates placement in the reply structure. Green posts initiate new threads (i.e., they are top-level threads), yellow posts reply directly to green posts, orange posts reply to yellow posts, and the pink post replies to the orange post.

In Figure 10.1, the author of each message and the time of its post are indicated. Users may post to multiple threads (e.g., Beth) or multiple times within a thread (e.g., Cathy). • Permanence: In many threaded conversations including email lists, once a message has been posted it cannot be rewritten or edited. A new message may be posted, but no matter how much someone may wish it, an original post often cannot be retracted. However, in many social media threaded conversations such as Facebook and LinkedIn, original posts can be modified or deleted after the initial contribution. • Thread Navigation: Threaded conversation systems also differ in how users navigate through the different threads. The partitioning of threads and messages into topics is a feature shared by many discussion interfaces. Most systems sort threads and messages in chronological or reverse chronological order (e.g., Figure 10.1). Other systems display threads or messages based on user ratings (e.g., Stack Overflow; Reddit). Often the aggregated views of the threads are the same for all users, but sometimes they are personalized for individuals, such as when Facebook's algorithms decide which threads to display on a feed, or unread messages are shown at the top of an interface. In addition to this basic structure, there are a few important ways that threaded conversation platforms differ. Some, such as email lists, are push technologies that send updates to all subscribers. Others, such as discussion forums and social media sites, are pull technologies that require individuals to visit a website in order to view messages. These are often accompanied by smartphone or email notifications when there are updates. Also, this chapter focuses on asynchronous threaded conversations, but many synchronous conversations such as texting, Instant Messenger, and group chat follow a similar reply structure even if the pace of interaction and nature of messages is different. Another important distinction is who can access content. Public conversations allow anyone who visits a website (or email list archive) to read the content, even if they are not part of the community. These are often indexed by search engines, helping their content rise out of obscurity. Semi-public conversations require users to create a username and log in, or join a group (e.g., on Facebook) before accessing content. While often anyone can join and gain access to prior messages, the content may not be indexed by search engines, making it more obscure. Finally, private conversations are only open to those who receive invitations via some existing member or are a member of some organization. Many corporate forums or email lists fall into this category.

III.  Social media network analysis case studies



10.3  What questions can be asked

The history of public online, threaded conversation communities began in the late 1970s with significant ­developments in the 1980s as bulletin board systems (BBS), email lists, and Usenet gained traction. The earliest threaded conversation systems relied on dialup connections, which encouraged groups within local telephone calling distance to form. Users of these early community systems demonstrated that text-only communication was sufficient to develop surprisingly meaningful relationships and rich cultures. Early communities covered topics ranging from dentistry to gaming to the occult. Access to public communities was often free. Others charged subscription fees, such as the WELL, a community of writers primarily from the San Francisco Bay Area described in Howard Rheingold's classic book “The Virtual Community: Homesteading on the Electronic Frontier” [2]. Technologies such as Listserv began to develop in the mid-1980s, allowing interested groups to create their own community email lists with increasing ease. As the Internet and the World Wide Web became ubiquitous in the 1990s, many of the original BBS services became or were replaced by Internet service providers (ISPs). Although they provided many services, asynchronous threaded conversation became one of the mainstays. Tools like Usenet that had relatively few users in the 1980s experienced exponential growth in the 1990s, growing from approximately 2,000 newsgroups in 1991 to close to 11,000 newsgroups in 1996. Today, email lists and discussion forums continue to support numerous communities ranging from neighborhood lists for sharing free items (e.g., Freecycle) to gaming communities, to medical support groups. Although email lists may sound passe, they can outperform social media as a marketing platform, since there is often less competition in one's email list than social media feed. These traditional threaded conversation platforms have proven surprisingly robust, enabling user groups to adapt them to a wide range of usage scenarios. Threaded conversations have also worked their way into a range of social networking sites, corporate intranets, multimedia sites (e.g., YouTube), customer review sites, Q&A sites, and specialized online community software. Reddit, with 18 billions page views per month in November 2018, and the wildly popular Stack Overflow have shown the power of combining threaded conversation with a voting-based navigation scheme. New mobile apps, such as Marco Polo support threaded video messaging, and popular social networking sites like Facebook include threaded conversations in their groups, fan pages, and profile pages. Corporate communication tools, such as Slack have integrated threaded conversation into a group chat system. Research on communities that use threaded conversation began in the early days of BBS and Usenet. Many

151

of the same themes continue to be explored today. For example, Kollock and Smith's 1999 book “Communities in Cyberspace” included chapters on identity online, deviant behavior and conflict management, social order and control, community structure and dynamics, visualization, and collective action [3]. All of these topics are still being explored in new contexts and with new technologies such as social networking sites, blogs, microblogging, and wikis. Early books by Preece [4], Kim [5], and Powasek [6] provided some enduring, practical advice and inspiration for those managing online communities. More recent additions by Kraut & Resnick [7] and Howard [8] continue the conversation on how to build community using online conversations and related technologies. Researchers from a variety of disciplines analyze threaded conversation communities and publish results in communication, business, information science, health, sociology, and computer science journals and conferences. Several have emerged around the Internet, such as the International Conference on Web and Social Media (ICWSM), Association for Internet Research (AoIR), the Journal of Computer Mediated Communication (JCMC), the Association for Computing Machinery's ComputerHuman Interaction (ACM-CHI), and ComputerSupported Cooperative Work (ACM-CSCW) conferences are just a few. Findings show that there is a consistent pattern of participation with few core members contributing the majority of content, many peripheral members contributing infrequently, and a large number of lurkers [9] who benefit by overhearing the conversations of others [10]. The nature of computer-mediated conversations depends largely on the type of community that engages in them. For example, technical and medical support communities differ in the level of empathy expressed [11] and the reusability of their content [12].

10.3  What questions can be asked There are many reasons to explore networks that form within large collections of conversations. New employees or community members need to rapidly catch up with the “story so far” to get to a point that they can make useful contributions. Community managers need tools to help them serve as metaphorical fire rangers and game wardens for huge populations of discussion contributors and the mass of content they produce. When outsiders such as researchers or competitors peer into a set of relationships, social network analysis can point out people, documents, and events that are most notable. A few of the specific questions that can be addressed with network analysis of community conversations are described next:

III.  Social media network analysis case studies

152

10.  Thread networks: Mapping message boards and email lists

• Individuals. Who are important individuals within the community? For example, who are the question answerers, discussion starters, and administrators? Who are the topic experts? Who would be a good replacement for an outgoing administrator? Who fills a unique niche? • Groups. Who makes up the core members of the community? How interconnected are the core group members? Are there subgroups within the larger community? If so, how are the subgroups interconnected? How do they differ? • Temporal comparisons. How have participation patterns and overall structural characteristics of the community changed over time? What does the progression of an individual from peripheral participant to core participant look like and who has made that transition well? How is the community structure affected by a major event like a new administrative team, the leaving of a prominent member, or an initiative to bring in new members? • Structural patterns. What network properties are related to community sustainability? What are the common social roles that recur among community members (e.g., answer person, discussion starter, questioner, administrator)?

10.4  Threaded conversation networks The network most commonly used to analyze threaded conversations is the Reply network. Each time someone replies to another person's message, she creates a directed tie to that other person. If she replies to the same person multiple times, a stronger weighted tie is created. To understand the nuances of how a Reply network gets created, you can compare the original data in Figure 10.1 to the Reply network derived from that data and shown in Figure 10.2. A related, but different network is the Top-Level Reply network, which connects all repliers to the person who started each thread (Figure  10.3) instead of the person they are replying to directly. This network emphasizes those who start threads (i.e., post the top-level message), while de-emphasizing conversations that occur midway through a thread. In some communities with short threads where all replies are typically directed at the original poster, such as Question and Answer (Q&A) sites or Reddit posts, this network can better reflect the underlying dynamics. However, in discussion communities or forums with longer threads, the standard Reply network is typically preferred because people later in the thread are often replying to each other.

FIGURE 10.2  An example discussion Reply network graph displayed in NodeXL, based on the data found in Figure 10.1. The network is constructed by creating an edge pointing from each replier to the person he or she replied to and then merging duplicate edges. Notice that Beth has replied directly to Dave twice, so the edge connecting them is thicker. Fiona replied to her own message, so there is a self-loop. Greg started a thread but was not replied to. He would normally not show up on the graph because he is not in the edge list; however, he was manually added to the Vertices tab and his visibility was set to Show, so he would appear.

III.  Social media network analysis case studies



10.4  Threaded conversation networks

153

FIGURE 10.3  A Top-Level Reply network graph displayed in NodeXL based on the data found in Figure 10.1. The network is constructed by creating an edge pointing from each replier to the person who started the thread (i.e., posted the top-level message) and then merging duplicate edges. Notice how Cathy plays a more prominent role (i.e., has a higher in-degree) than in the standard Reply network graph (Figure 10.2) because she started the longest thread and all subsequent repliers link to her. Self-loops are more frequent in this type of network because people like Cathy may respond to those who replied to her initially, leading to a self-loop.

Affiliation data connecting posters to threads (or forums) can also be used to create bimodal networks (see Chapter  6). These are undirected, weighted networks that connect posters (i.e., users) to threads. For example, an edge would connect Cathy to Thread B with a weight of 2 because she posted to that thread twice. Beth would be connected with a weight of 1 to Thread A, Thread B, and Thread C because she posted to each of them once. This network is ideal for identifying boundary spanners, as you will see in Section 10.6. As discussed in Advanced topic: Transforming a bimodal affiliation network into two unimodal network of Chapter 6, affiliation networks can be transformed into two additional undirected, weighted networks. With threaded networks, you can create a user-to-user network connecting people based on the number of threads (or forums) they both contribute to, and a thread-to-thread (or forum-to-forum) network connecting threads together based on the number of contributors they share. These networks are good for creating overview graphs of large communities with many threads or forums. Preparing data needed to create threaded conversation networks can be challenging because they rely on such a wide range of technologies. Email lists are the easiest conversations to capture in NodeXL, because you can use the email import wizard (see Chapter  9). Data

from discussion forums, Reddit, Facebook, etc. must be generated by using screen scrapers, manually entering data, or performing queries on the forum's database or through a web Application Programming Interface (API). Whichever approach you use, your dataset will likely have header information that includes some of the following information for each message: a time stamp, a message author, an identifier for the message this message is a reply to (if any), a subject line (or thread ID), a set of tags, an attachment, a link to the author's profile, a group or forum the thread is a part of, and a rating. A separate file of information on each user is also often useful. It may include aggregate participation statistics on other community activities (see Section 10.6 for an example). All of this data can be useful in creating maps of conversation networks, but at the core a simple edge list is the minimum necessary requirement to start a social network analysis of a conversation. The type of discussion platform in use will also influence the potential data problems you are likely to run into. For example, email lists often have people registered with multiple email addresses, making it necessary to combine duplicate addresses for the same person (see Chapter 9). Email lists also have problems when the reply structure isn't clear, because people reply to the email list address rather than to one another. Corporate

III.  Social media network analysis case studies

154

10.  Thread networks: Mapping message boards and email lists

email lists and discussion forums typically have the cleanest set of unique identifiers for individuals, but even then, name and title changes can cause problems. It is also important to realize that the reply network from an email list only captures messages sent or posted to the list. Many personal messages sent directly to and from individuals on the email list are not captured. Depending on the type of community and the default settings (e.g., email list Reply To settings), these private messages may account for the majority of all messages exchanged among a population. In addition, other types of communications such as corporate meetings, phone calls, and instant messenger exchanges are invisible to discussion forum networks. People who communicate and contribute more effectively through other channels may show up only marginally in discussion forum ­datasets. However, even given these limitations, analysis of threaded conversation networks can provide vital information about community dynamics and help identify important individuals and groups.

10.5  Identifying important people and social roles in the CSS-D Q&A reply network There are a host of email lists, forums, and Q&A websites such as Stack overflow and Quora where people post technical questions and volunteers provide answers. Many companies host these Q&A discussions to learn about problems with existing products, resolve customer concerns, generate new ideas on future improvements, and build a loyal customer community. To meet these goals, it is often important to understand which individuals play important roles within the community, something that can be challenging when managing multiple, active communities or viewing content from across large sites. In this section you will learn how to identify key members of a technical support Q&A discussion focused on cascading style sheets (CSS), which are used in front-end web development. During the time of data collection, the community sent approximately 50 messages a day and included several key administrators who kept the conversation friendly and on topic. See [12] for a complete description of the community and some of the strategies used by the community administrators and members to make it so effective. In an online community, users contribute in different patterns and styles. In other words, community members fill different social roles. Understanding the composition of social roles within your community or social networking site can provide many insights that can help you be a more effective community manager. Social network analysis provides metrics that can be used to

a­ utomatically identify those who fill unique social roles and track their prevalence over time. This can help community and social media managers: • Identify high-value contributors of different types. Which community members are the most important question answerers or question starters? Who connects many other users together? Answering these questions can help you know who to thank (and for what) and how to support individuals' needs. • Determine if your community has the right mix of people. Is your community attracting enough question answerers? Are there enough connectors to hold the community together? Is discussion crowding out Q&A? Is a discussion space dissolving into Q&A? Knowing the answers to these questions can help you know who to recruit or encourage more, as well as what policies may be needed. • Recognize changes and vulnerabilities in the social space. How has the community composition changed as it has grown? What is the effect of a certain prominent member leaving the community going to have? Which members are currently irreplaceable in the type of work they do? What is the effect of a policy change or change in settings on the community dynamics? Answering these questions can help you prepare for change, understand the effects of prior decisions and events, and cultivate important relationships. In this section you will learn how to identify important individuals and social roles within the CSS-D community. You will do this by using subgraph images (introduced in Chapter 7) and creating a composite metric that helps identify the two most important social roles within Q&A communities like CSS-D: answer people and discussion people. You will then use this metric to develop visualizations that show the relationships between these individuals. The easiest way to get a sense of the key individuals within a network is to create the 1.5 subgraph images for each vertex using a layout like the Harel-Koren Fast Multiscale layout (see Figure 10.4). Once these subgraph images of each email contributor's local networks are created, you can sort on the graph metrics in the Vertices worksheet associated with each contributor such as in-degree (who receives messages from the most people) and out-degree (who sends messages to the most people) to bring differently connected individuals to the top. You can also sort by centrality measures like Page Rank to get a sense of who is a core member of the community, because this member is an active participant and talks to other active participants.

III.  Social media network analysis case studies



10.5  Identifying important people and social roles in the CSS-D Q&A reply network

155

Question People •

Low In- and Out-Degree

•

High Avg Degree of Neighbors

Answer People •

High % Out-Degree

•

Low Clustering Coefficient

•

Low Avg Degree of Neigbors

Discussion Starters •

Low % Out-Degree

•

High Clustering Coefficient

•

High Avg Degree of Neighbors

FIGURE 10.4  NodeXL Subgraph images (1.5 degree; vertex and incident edges highlighted red) for six CSS-D contributors that fill three different social roles within the CSS-D community. Answer people predominantly reply to questions from isolates (i.e., those who are not connected to others). Question people typically have a low degree themselves, but they receive messages from those with high degree (i.e., answer people). Discussion starters initiate long threads and receive many replies, often from people who know each other.

Scanning through the Subgraph Images of CSS-D contributors helps you get a sense of the different social roles that exist within the email list community. Figure  10.4 shows examples of three types of contributors (question people, answer people, and discussion starters) along with some of the metrics that could be used to identify them (see Advanced topic: Social role measures for more). Question people post a question and receive a reply by one or two individuals who are likely to be answer people. Answer people mostly send messages (arrows point toward other vertices) to individuals who are not well connected themselves [13]. Discussion starters mostly receive messages (arrows point toward them), often from people who are well connected to each other. You can typically identify a person's social role by looking at his or her subgraph image (Figure 10.4), but doing so for many individuals becomes problematic.

Instead, it is possible to create aggregate metrics that automatically identify those who play certain social roles. These metrics consist of a combination of network metrics and participation metrics. Automatically identifying social roles within a community using metrics facilitates their tracking over time, which allows you to keep your pulse on the health of your community. It can also be used in combination with visualizations as shown in Figures  10.5 and 10.6 to more easily understand individuals' social roles and how they relate to one another. The specific metrics used to identify social roles will depend on the metrics that are available (i.e., those that are tracked) and will be tied to some extent to the underlying type of social media being analyzed. Advanced topic: Social role measures describes several metrics that can be used to identify different roles within threaded conversations.

III.  Social media network analysis case studies

FIGURE 10.5  NodeXL map of the CSS-D Q&A network after removing the vertex for the email list address itself. Answer people (greener) and discussion starters (redder) are identified by the calculated answer person score (see Advanced topic: Social role measures). Blue vertices have a total degree of fewer than 15. Subgraph images (1.5) of the top four discussion starters are shown. Vertex size is mapped to eigenvector centrality. Edge weight is mapped to both edge size (1.5–4) and opacity (20–80), applying the logarithmic scale and ignoring outliers. Like many help-based communities, CSS-D consists of mostly question askers with a handful of answer people and discussion starters.

FIGURE 10.6  NodeXL map of the filtered version of the CSS-D email list seen in Figure 10.5 showing only the most central members. The maximum size of vertices and edges has been increased to more clearly draw comparisons.

III.  Social media network analysis case studies



10.5  Identifying important people and social roles in the CSS-D Q&A reply network

157

A D VA N C E D T O P I C Social role measures

Custom metrics can be created based on network or attribute data. Some metrics, such as the core social network metrics, are created automatically using the Graph Metrics feature (see Chapter  6). Other metrics, such as the average posts per thread or the days active, must be tracked via some other means. All of the metrics presented in this section are devised so that higher values correspond with typical answer person behaviors. Depending on what data you have available, you can combine different metrics into an aggregate metric by averaging them, multiplying by them, or taking a weighted average. Table 10.1 shows a list of different custom metrics that can be created, alongside their description and interpretation. Figures 10.5 and 10.6 show values for a Degree_Cuttoff, Percent_Out-Degree, and Ans_Person_Score (i.e., answer person score) that are used in the visualizations. The Degree_Cuttoff equals Out-Degree + In-Degree, so that those with a low total degree can be filtered out (e.g., those with under 15). The Percent_Out-Degree is described in the last row of Table  10.1 and is used to identify those who reply to others, versus have people reply to them. Finally, the Ans_Person_Score is calculated by multiplying the Percent Out_Degree by (1 − Clustering Coefficient), which indicates that the person has both a high Out-Degree, as well as a low number of people they message who message each other. Thus, high values of the answer person score identify answer people. In addition, low values identify discussion starters, as they have a high percent in-degree

The specific social roles and their prevalence within a particular community will depend on the nature of that community. Because the CSS-D community is primarily a Q&A community, it consists of mostly question askers, a handful of prominent answer people, and a small number of discussion starters. Other more discussion-based communities would have many more discussion starters as well as other social roles such as flame warriors, commentators, and connectors. Tracking the ratio of people that play different social roles can be a good way to assure that a community is healthy. For example, if the CSS-D community had too few answer people or an influx of many question people, it could not function effectively. Viewing the entire reply network for the CSS-D email list (Figure  10.5) can provide some general insights about the composition of its population, although the size of the network can make it challenging to interpret without filtering. Figure  10.5 maps the answer person score (see Advanced topic: Social role

(they solicit replies from many people while not sending messages to many people) and high clustering coefficient (those who they are connected to know each other). TABLE 10.1  Social role metrics. Metric

Description

(User's Thread Count) ÷ Brevity is preferred. Larger values = fewer messages per thread (User's Post Count) (User's Reply Posts) ÷ (User's Total Posts)

Initiation is avoided. Larger values = avoids starting threads

(User's Degree) ÷ (Total Users)

Talks to many people. Larger values = replies to a significant fraction of community members

(1 − Clustering Coefficient)

Talks to people who aren't well connected to each other. Larger values = lower clustering coefficient (i.e., less well-connected neighbors)

1 ÷ Avg of Neighbor's Degree

Talks to people who connect to few others. Larger values = talks to more isolates

(User's Days Active) ÷ (User's Possible Active Days)

Posts on most days. Larger values = posts on multiple days more often

(User's Out-Degree) ÷ (User's Out-Degree + User's In-Degree)

Percent out-degree. Larger values = is connected to more people because of replying to them than because of receiving from them

measures) to color: green-colored nodes represent answer people, r­ed-­colored nodes represent discussion starters, and blue nodes have a total degree of less than 15. Larger nodes have a higher eigenvector centrality suggesting they are connected to many people and others who are well connected. The binned layout is used to identify isolates, of which there are many because the email list address itself was removed. Isolates represent those who posted to the list and didn't receive a response (e.g., they posted an announcement) or in some cases those who replied to the list without copying in the address of the person who they were replying to. Overall the composite network shows many individuals connected primarily through a handful of central question answerers and a small but stable core group of members that interact with one another regularly. To better focus in on the core members of the community you can filter out vertices with a total degree of less than 15. Figure  10.6 shows the resulting network a­ fter

III.  Social media network analysis case studies

158

10.  Thread networks: Mapping message boards and email lists

manually positioning the vertices. Subgraph images for the top three answer people are shown. The edge weights, mapped to the edge width and opacity, provide a good sense of who interacted with whom during the 2-month time period and is thus likely to know each other and perhaps have similar interests. Note that even among these core members, discussion starters rarely reply to other discussion starters. Also notice that the largest vertex (i.e., the one with the highest eigenvector centrality), while categorized as an answer person, receives many messages from the core members. This suggests that he plays multiple important roles within the community. In fact, if he were removed from the network, there would be considerably fewer connections between the core members. This suggests that community administrators should make sure this individual is adequately appreciated and encouraged to remain in the community.

10.6  Understanding groups at Ravelry Ravelry (www.ravelry.com) is a thriving online community for anyone passionate about yarn. Millions of knitters and crocheters have registered on the site. Users organize their projects, yarn stashes, and needles; share and discover designs, ideas, and techniques; and form friendships through discussions and exploration of shared interests. In this section, you play the role of a fictional Ravelry community administrator. You will work with data on the top 20 posters to three discussion forums created for different groups. The data and initial network analysis for this section were developed by Rachel Collins, a graduate student at University of Maryland's iSchool. Special thanks to the Ravelry community for allowing us to analyze it and discuss their fascinating community in the book. All group and individual usernames have been modified for privacy reasons. The techniques used to analyze this bi-modal network can be applied to many similar networks that connect people to discussions or other items. Imagine you are assigned three group discussion forums to monitor and help develop. They are highly active groups, making it hard to keep up with all the messages and see the forest from the trees. You'd like to get a better sense of how the most important community members relate to one another, as well as how the groups differ. This understanding will help you recommend the best group for a newcomer to join (which could mean you link to the forum more prominently in your website), as well as identify individuals with certain expertise or social relations that you can call upon if needed. You also hope to share the visualization with the groups themselves to encourage self-reflection.

The three groups you are in charge of include one common-­ interest group (Apathetic Funloving Crafters [AFC]), one meetup (Chicago Fiber Arts), and one knitalong (Project Needy). They are three of hundreds of similar groups. Discussion forums for each group serve as their central hubs. Individuals can participate in as many forum groups as they desire. You have collected data on project output, discussion board usage, blog activity, community roles, and total friends for the top 20 posters in each group. This lets you relate many different activities together in a single analysis, focusing attention on the most active members who are typically the most important. Figure  10.7 shows a bimodal graph of the three forums/groups (shown in blue) connected to individuals who have posted to them. Edge thickness is based on the number of forum posts (using a logarithmic mapping). The thinnest lines connect users to groups that they are members of but have not yet posted to. Other visual properties are used to convey individuals' levels of activity in other parts of the community. Maroon vertices also maintain a community blog, whereas solid disks are community moderators or volunteer editors. The graph helps you identify important individuals, such as those who post to multiple groups or have certain color/size/shape combinations. It also helps you compare the three groups. For example, the graph makes clear that the Apathetic Funloving Crafters (AFC) forum is very active, includes many bloggers, and includes relatively few people who complete a large number of projects (perhaps explaining the “Apathetic” in the title). In contrast, the Project Needy group includes many highly productive members, many of whom are both administrators and bloggers. In contrast, the Chicago Fiber Arts group has fewer bloggers and less project activity. A newcomer to the Ravelry community could use a visualization like the one displayed in Figure  10.7 to quickly get a sense of which group(s) he or she may want to join, as well as identify some of the prominent members. Administrators can use similar graphs to identify potential candidates for volunteer editors or identify clusters of boundary spanners with which to form new groups because of shared interests. Providing graphs like this one to the groups themselves can also prompt self-reflection and potentially foster new connections. They can also be used to better understand how the activities on the site relate to one another, although use of statistics may be needed to more systematically validate initial claims. For example, Figure 10.7 shows that location-based groups have a lower percentage of active members who blog and people who complete many projects seem to cluster into project groups. Understanding trends such as these can help you better target your community and groups around the different user types involved.

III.  Social media network analysis case studies



159

10.8  Researcher's agenda

FIGURE  10.7  Bimodal network connecting three Ravelry groups (i.e., forums) represented as blue text boxes to contributors represented as circles. Edge width is based on number of posts (with logarithmic mapping). Vertex size is based on number of completed Ravelry projects. Maroon vertices have a blog and solid circles are either community moderators or volunteer editors. The network helps identify important boundary spanners (e.g., those connected to multiple groups) as well as compare groups.

10.7  Practitioner's summary Many online communities use threaded conversations in the form of email lists, discussion forums, Facebook groups, Reddit, and Q&A sites like Quora and Stack Overflow. Although they make use of a wide range of technologies that differ in their delivery infrastructure, all threaded conversations share similar characteristics: they are composed of single-authored messages organized into threads (i.e., a top-level message, replies to that message, and possibly replies to those replies), threads are often found within topics or groups, messages are often permanent, and users often have a shared view of the conversation. These conversations lend themselves to the creation of several networks including the directed, weighted Reply network and TopLevel Reply network; the undirected, weighted affiliation network connecting threads (or forums) to the individuals that posted to them; and the undirected, weighted networks

derived from the affiliation network including the user-touser network and thread-to-thread network. The analysis of the CSS-D technical support community showed how to identify important social roles and individuals who fill those roles including answer people, discussion starters, and questioners. The analysis of Ravelry forums and posters showed how to use a bimodal affiliation network to understand how forum-based groups are connected, identify important boundary spanners, and relate nondiscussion network metrics (e.g., blog activity, project activity) to group discussion activity.

10.8  Researcher's agenda Research on threaded conversation communities has a long history as outlined in Section 10.2, yet there ­remain many interesting research questions to explore.

III.  Social media network analysis case studies

160

10.  Thread networks: Mapping message boards and email lists

As threaded conversations become embedded within more complex social spaces with multiple interaction technologies, it is increasingly important to understand how they all interact. For example, Hansen has found that technical and patient support groups benefit from combining a threaded conversation (i.e., email list) with a more permanent wiki repository [12]. The Ravelry example showed strategies that have not yet been widely used by the research community to understand how network position relates to use of other tools (i.e., blogs) or activities (i.e., projects). Network-based research is also needed to better understand the determinants of successful online communities. For example, we don't know what proportion of mixtures of answer people, discussion starters, and questioners lead to better outcomes or what overall network statistics (e.g., clustering coefficient) are correlated to success. From a design perspective, there are many fascinating opportunities to enhance the threaded conversation model [1] as evidenced by new features on sites like Quora, Reddit, and Stack Overflow. One particularly promising approach is to use visualization to help people make sense of community interaction [14, 15].

References [1] P. Resnick, D. Hansen, J. Riedl, L. Terveen, M. Ackerman, Beyond threaded conversation, in: CHI ‘05 Extended Abstracts on Human Factors in Computing Systems (Portland, OR, USA, April 02–07, 2005). CHI ‘05, ACM, New York, NY, 2005, pp. 2138–2139. [2] H.  Rheingold, The Virtual Community: Homesteading on the Electronic Frontier, Adison-Wesley Pub. Co, Reading, MA, 1993. [3] M.  Smith, P.  Kollock (Eds.), Communities in Cyberspace, Routeledge, London, UK, 1999. [4] J. Preece, Online Communities: Designing Usability and Supporting Sociability, John Wiley & Sons, Inc, New York, NY, 2000. [5] A.J. Kim, Community Building on the Web: Secret Strategies for Successful Online Communities, first ed., Peachpit Press, 2000.

[6] D. Powazek, Design for Community, illustrated ed, Waite Group Press, 2001. [7] R.E. Kraut, P. Resnick, Building Successful Online Communities: Evidence-Based Social Design, MIT Press, 2012. [8] T.  Howard, Design to Thrive: Creating Social Networks and Online Communities that Last, Morgan Kaufmann, 2009. [9] B. Nonnecke, J. Preece, Lurker demographics: Counting the silent, in: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (the Hague, the Netherlands, April 01–06, 2000), CHI’00. ACM, New York, NY, 2000, pp. 73–80. [10] D.L.  Hansen, Overhearing the Crowd: An Empirical Examination of Conversation Reuse in a Technical Support Community, in: Proceedings of the Fourth International Conference on Communities and Technologies (University Park, PA, USA, June 25–27, 2009). C&T’09, ACM, New York, NY, 2009, pp. 155–164. [11] J. Preece, K. Ghozati, In search of empathy online: A review of 100 online communities, in: Proceedings of the 1998 Association for Information Systems Americas Conference, 1998, pp. 92–94. [12] D. Hansen, Knowledge Sharing, Maintenance, and Use in Online Support Communities, Unpublished Dissertation, University of Michigan, http://hdl.handle.net/2027.42/57608. [13] H.T. Welser, E. Gleave, D. Fisher, M. Smith, Visualizing the signatures of social roles in online discussion groups, J. Social Struct. 8 (2) (2007). [14] Y. Chen, Visual opinion analysis of threaded discussions, in: Data Mining Workshop (ICDMW), 2015 IEEE International Conference on, IEEE, 2015, November, pp. 646–651. [15] F.B. Viégas, M. Smith, Newsgroup crowds and AuthorLines: Visualizing the activity of individuals in conversational cyberspaces, Proceedings of Hawaii International Conference on Software and Systems (HICSS) 2004. [Best Paper: Persistent Conversation Minitrack]

Further reading B. Butler, L. Sproull, S. Kiesler, R.E. Kraut, Community effort in online groups: Who does the work and why, in: S. Weisband, L. Atwater (Eds.), Leadership at a Distance, Lawrence Erlbaum Associates Inc, Mahwah, NJ, 2005. E. Wenger, Communities of Practice: Learning, Meaning and Identity, Cambridge University Press, Cambridge, 1998.

III.  Social media network analysis case studies

C H A P T E R

11 Twitter: Information flows, influencers, and organic communities O U T L I N E 11.1 Introduction

11.6.1 User-level visual properties 173 11.6.2 Cluster-level layout and visual properties 174

161

11.2 Defining your topic-networks: Formulating a social media monitoring query 161 11.3 Twitter data collection

162

11.4 The raw data layout

163

11.5 Network analysis 11.5.1 Vertex-level metrics 11.5.2 Network-level metrics 11.5.3 Groups

166 166 169 170

11.6 Visualization

172

11.1 Introduction In social media, users form social networks by making relationships with other users. On Twitter, users make relationships when they follow other users, mention users in a tweet, reply to a tweet, or “retweet” another tweet. Once you follow another user by subscribing to their posted content, this content will then appear on your Twitter feed. This can be thought of as an ­awareness connection. Mentioning another user in a post is an indication of attention and therefore exposure to content. A retweet is a key mechanism on Twitter for amplifying content. The collection of the relationships formed by these activities create social networks. In this chapter you will learn to collect, analyze and visualize topic-specific Twitter conversation networks. You will first use NodeXL's Twitter Importers to collect Twitter network data—the content of a set of Tweets, and the relationships and user-information that can be extracted from them. Twitter data can be collected about any topic of your interest, from archery to zebras. This chapter will take you step-by-step through the process

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00011-X

11.7 Analysis of content

175

11.8 Share your work on the NodeXL graph gallery

177

11.9 Practitioner's summary

177

11.10 Researcher's agenda

177

References

178

Suggested reading

178

of calculating and interpreting key social networks metrics, at the user, group and network levels. It builds upon and extends the analysis of the CSCW 2018 Twitter network explored in Chapters 6 and 7, which should be read beforehand. You will conclude this chapter by visualizing Twitter networks and highlighting key findings. Identifying the prominent users who can influence large groups or understanding the polarized discussions will help you to intervene successfully, avoid conflicts, and understand the roles of competing groups.

11.2  Defining your topic-networks: Formulating a social media monitoring query Twitter is not explicitly organized into topic specific discussions, like discussion forums (Chapter  10). Instead, individual users can tweet about a wide range of personal, political and social topics. On Twitter, the only indication of a tweet's topic is found within its content, which includes hashtags or keywords relevant to the topic. As a result, discussion “communities” are

161

© 2020 Elsevier Inc. All rights reserved.

162

11.  Twitter: Information flows, influencers, and organic communities

much more dynamic and emergent. Subgroups of users are selected based on their use of a particular keyword. For example, the relationships of all users who mentioned “HPV” (short for human papilloma virus) during a particular period of time creates a dataset containing a slice of the HPV topic-network. You can use NodeXL to collect, analyze and visualize this type of social media network data from Twitter. Twitter conversations span a wide range of topic and issues. In order to collect topic-specific Twitter network data, you must first define your topic. State your topic first (e.g., Soft drinks and obesity, or the name of a political party, celebrity, or product). Next, explore related keywords and hashtags using the built-in Twitter Search and third party apps such as https://hashtagify.me to identify trends, popularity, and related terms. In order to communicate your topic to Twitter, you will use a Boolean search query that combines these terms into a specific machine-readable format. A Boolean search is a query technique that utilizes Boolean Logic to connect individual keywords or phrases within a single query. The term “Boolean” refers to a system of logic developed by the mathematician and early computer pioneer, George Boole. Boolean searching includes three key Boolean operators: AND, OR, and NOT. • An AND operator narrows your search. Between two keywords it results in a search for posts containing both of the words. For instance, the Boolean search “Cats AND Dogs” will retrieve all posts that contain both words. A lack of an operator between words is interpreted as an AND operator within NodeXL. For example: “Cats Dogs” is treated the same as “Cats AND Dogs”. Example returned post: “Vaccinate your dogs, cats and ferrets. Show your pets you care!” • An OR operator expands your search. An OR operator will return any posts that contain at least one of the search terms. For instance, the Boolean search “Cats OR Dogs” will retrieve posts that contain either word. Example returned post: “love all dogs!” • A NOT operator excludes posts containing the keyword. Using the NOT operator will exclude any posts containing the keyword following the operator. For instance, the Boolean search “Cats NOT Dogs” will retrieve all posts that have the word “Cats” in them, unless it also has the word “Dogs.” Twitter also recognizes the minus sign (−) as a NOT operator (E.g.” Cats -Dogs”) within Boolean searches. When collecting data using NodeXL you should use the “-” operator rather than the word NOT. This post will be included: “I love cats!” This post will not be included: “I love cats and dogs!”

Two other indicators that are helpful in constructing a Boolean search are parentheses and quotation marks. • Quotation marks requires words to be searched as a phrase, in the exact order you type them. For example, searching for posts about the movie Love Actually, using the quotation marks will search for the phrase exactly as it appears (i.e., “Love Actually”). This post with be included: “Hugh Grant is a great fit for his role in Love Actually.” This post will not be included: “Actually, falling in love is not as simple as you may think.” • Parentheses require the terms and operations that occur inside them to be searched first. Sometimes called nesting, parentheses add a level of organization for your Boolean search, allowing you to formulate complex search strings. Consider the following Boolean search: “(Cats OR Kittens) AND (Dogs OR Puppies).” The search will first be performed within the parentheses, and only then between them. Results would include any post that contains a combination of the two nested boolean queries. These post will be included: “I love all kittens and puppies; I love dogs and cats.” This post will not be included: “I love cats and especially kittens.” You can learn more about Twitter search operators on the Twitter search page (https://twitter.com/searchhome). Check out the operators and advanced search links. Once you construct a Boolean search to collect your topic network you can use the Twitter search page to test your search string (https://twitter.com/search-home). You can learn more about search options by clicking on the operators link in the Twitter search page. Include keywords, pairs of words, relevant hashtags, and handles as appropriate. Revise and retest your search result, until you you feel that it captures the data you need. Once you feel your search string properly captures the tweets you want to analyze, it helps to make a copy of the string for future reference.

11.3  Twitter data collection Open NodeXL Pro and click on the NodeXL Pro tab along the top menu ribbon in Excel. Click Import in the Data section of the NodeXL ribbon. In the drop-down menu, choos From Twitter Search Network as shown in Figure 11.1. You can use the NodeXL Twitter Search Network importer dialog (see Figure 11.2) to extract networks for your search string. Several options are available to refine data requests using this importer. The NodeXL data c­ ollector starts by performing a query against the Twitter Search

III.  Social media network analysis case studies



163

11.4  The raw data layout

•

• •

•

• •

• •

FIGURE 11.1  Data import sources for NodeXL.

service at http://search.twitter.com. Searches can be performed for any string of characters. This service returns up to 18,000 tweets that contain a requested search string. • Enter your search query in the Search for tweets that match this query field. • Select Basic network. This option will collect tweets associated with your search term and generate a network based on their Mention and Reply-to relationships. The second option collects the Follow relationships, however it is very slow due to data restrictions imposed by Twitter and should only be used for very small datasets. • If it is the first time you are using NodeXL to collect Twitter data, select I have a Twitter account, but I have not yet authorized NodeXL to use my account to import Twitter networks. Take me to Twitter's authorization page. NodeXL will send you to Twitter to login to your account. When you grant NodeXL access, Twitter will provide you with

an eight digit code. Copy it and paste it back in NodeXL in a box that will pop up. After the one-time authorization process, the second option (I have a Twitter account, and I have authorised NodeXL to use my account to import Twitter networks.) will be selected. Set the maximum number of tweets you wish to collect. By default, this value is set to 18,000 tweets, the maximum allowed by Twitter's API limits. Make sure that Expand URLs in Tweets is selected. NodeXL will expand the commonly used shortened URLs (e.g., tinyurl.com, ow.ly, etc.) to make sure they appear in their original long form. Click OK to tell NodeXL to begin collecting data. This can take a long time because of limits in the rate of data delivery imposed by Twitter. It can take up to an hour to collect a very large dataset. When data collection is complete, you may be asked to allow Excel to wrap text in columns. Choose Yes. You may ask yourself “Do I have enough data?” There is no definitive answer, of course, as the dataset captures the recent Twitter conversation about your topic. For this exercise, it would be ideal if you have at least 1,000 vertices and 1,000 edges. Save your data. When naming the file, use indicators such as Twitter, date of data collection and the first few keywords in the query used to collect the data. Make a backup copy of your data. Use it for your analysis, so you can always return to the original raw data if needed. Just like in real life, there are no redos in NodeXL!

Alternatively, you may choose to work with the sample dataset that was used for this chapter: http://nodexlgraphgallery.org/Pages/Graph.aspx?graphID=172973

11.4  The raw data layout The raw Twitter data you have just downloaded will populate the Edges and Vertices worksheets. In this Twitter analysis, the Edges worksheet contains a row for each edge which represents a connection event between two people who tweeted within the collected dataset. Edges represent the various kinds of relationships that can be created through Twitter. NodeXL constructs four different types of Twitter edges: Mentions, Replies to, Retweet, and Tweet. A Mentions edge is created when one user creates a tweet that contains the name of another user, which is NOT at the beginning of the tweet (indicated with a preceding “@” character: “just spoke about social media with @marc_smith”). If more than one users are mentioned in a tweet, a new edge (e.g., row on the Edges worksheet) will be created for each mentioned user. For example, “just spoke with @marc_smith and @shakmatt about

III.  Social media network analysis case studies

164

11.  Twitter: Information flows, influencers, and organic communities

FIGURE 11.2  NodeXL data importer for Twitter search networks.

NodeXL” would create an edge pointing from the sender to @marc_smith and another from the sender to @shakmatt. A Replies to edge is created w ­ henever there is a username at the very beginning of a tweet (e.g., “@marc_ smith just spoke about social media”). If multiple people are mentioned in such a tweet, a Replies to edge is only created for the first person mentioned. A Retweet edge is created when a user reposts or forwards a tweet written by someone else. The Retweet ID column includes the ID of the original post and the Retweet Count column includes details on how many times it was retweeted. A Tweet edge is created when it does not fall into any of the other categories (i.e., no usernames are included and it is an original message). Tweet edges are “self-loops” with the same user linking back to themselves. Meanwhile, in the Vertices worksheet, each Vertex refers to a specific Twitter user who appears within the dataset. In the edges worksheet, then, a single tweet is converted into one or more edges. Each row in the edges worksheet represents a connection between two users (see Figures 11.3 and 11.4). • Vertex 1 and Vertex 2 represents the two users connected by a tweet. Vertex 1 is the author of the tweet

•

• • • •

•

and Vertex 2 is the target of the tweet by virtue of being mentioned, replied-to, retweeted, or tweeted (in which case the same username shows up in both columns). Note that Twitter networks are Directed networks. The Visual Properties and Labels columns are not part of the raw data, and therefore empty at the start of the analysis. You will learn more about these values after they are populated. The type of connection, Mentions, Replies to, Retweet, and Tweet, is identified in the Relationship column. Relationship Date refers to the date and time when the tweet was posted in Coordinated Universal Time (UTC), also known as Greenwich Mean Time (GMT-0). The Tweet column includes the text of the tweet itself. The URLs column lists the full expanded hyperlinks of any shortened links mentioned in the tweet text. This is created when the expand URLs in Tweets option is selected. The Hashtags in Tweets column extracts the hashtags from the tweet for further analysis.

The Vertex worksheet provides information about individual users in the network. Each row represents a

III.  Social media network analysis case studies



11.4  The raw data layout

FIGURE 11.3  The Edges worksheet encodes user relationships derived from analysis of each Tweet.

FIGURE 11.4  The Edges worksheet showing tweet detail information.

III.  Social media network analysis case studies

165

166

11.  Twitter: Information flows, influencers, and organic communities

FIGURE 11.5  The Vertex worksheet.

twitter user in the network. You will find vertex-level metrics, once calculated, in this worksheet. The Vertex worksheet also includes metadata information from the Twitter platform about users in the network. This includes the user's “handle” in the Vertex column; the number of users who follow them (i.e., Followers), the number of users they follow (i.e., Followed), and profile information such as Description, Location and Website (see Figure 11.5).

11.5  Network analysis 11.5.1  Vertex-level metrics Applying social network analysis to Twitter activity, users are characterized based on their connectivity in the network. Measuring how central users are in the network reveals the influential users and their connections to one another. Chapters 3 and 6 provide a detailed description of the metrics used in this chapter. Open the Graph Metrics dialog in the NodeXL Ribbon and

s­ elect the metrics you will use in this chapter, which are shown in Figure 11.6. In and out degree centrality Degree centrality metrics count the number of connections (edges) a user (vertex) has in the network. Because Twitter networks are directed (e.g., @Aviva may mention @Hans, but @Hans may not mention @Aviva), degree centrality can take two forms. In-degree centrality measures the number of edges others have initiated with a vertex. For instance, if @Aviva was mentioned 5 times by users in a Twitter topic-network, her in-degree centrality metric would be 5. Out-degree centrality counts the number of edge a vertex has initiated with others. If @Hans mentioned 10 other users in his tweet, his out-­degree centrality would be 10. In and out degree centrality have important implications for evaluating the importance of users in Twitter networks. Users with high in-degree centrality gain attention to their tweets among the community of users who participate in the conversation about that topic. Users' in-­degree centrality, thus, captures the community's engagement with them. Those with high in-degree centrality scores can

III.  Social media network analysis case studies



11.5  Network analysis

167

FIGURE 11.6  Network Graph Metrics chosen for the Twitter analysis performed in this chapter.

be thought of as conversational hubs, since others have mentioned, replied to, or retweeted their posts. In-degree centrality, therefore, is an indication of the cascades of information flow initiated by a user. Keep in mind that a user may have high in-degree in one topic-network and low ­in-­degree in another. For example, a food blogger may have high ­in-­degree centrality in a food-related topic network, but low in a car-related network. Out-degree centrality captures the outreach of a user to the community. A high out-degree centrality value indicates that a user tweets a lot about a topic, aiming to reach users' attention by mentioning or replying to them. Out-degree centrality, then, captures the level of engagement a user initiates with members of the community. In and out degree centrality metrics capture users' engagements with other users and their content. It indicates actual attention given to content and actions users

took to disseminate information. This is in contrast to the number of followers one has on Twitter, which only indicates the potential reach of a tweet. A high number of followers doesn't necessarily equate to all followers actually reading tweets, though it does indicate a certain level of popularity. Another important difference between the the degree and follow metrics is that the in and out degrees measure centrality within the topic-­ network, whereas following and follower metrics define centrality within the entire Twitter network, and are not topic-specific. It is often useful to differentiate between “local” conversation hubs (i.e., those with high in degree in a topical network) and “global” hubs (i.e., those with high number of Twitter followers). Both can be valuable to identify, but often for different reasons. Now look at the degree centrality values. Select the Vertices worksheet and find the in-degree centrality and

III.  Social media network analysis case studies

168

11.  Twitter: Information flows, influencers, and organic communities

FIGURE 11.7  Sort by value.

out-degree centrality metrics columns. Using the sort option in NodeXL (Figure 11.7), you can answer the following questions: • Who are the top in-degree users? • Who are the top out-degree users? • Who are the users with the most Twitter followers? Explore some of the other columns, such as description and web, to learn more about your top users. Betweenness centrality Degree centrality metrics define a vertex's centrality by number of connections in a network. A user (i.e., a vertex) may also be central in a network because it connects users that would otherwise be disconnected or less connected. Betweenness centrality measures the extent to which a vertex plays this bridging role in a network. Specifically, betweenness centrality measures the extent that the user falls on the shortest path between other pairs of users in the network (see Chapter 6). The more people depend on a user to make connections with other people, the higher that user's betweenness centrality becomes. Burt's [1] theory of structural holes examines social actors (e.g., individuals and organizations) in unique positions in a social network, where they connect other actors that otherwise would be much less connected, if at all. In Burt's [2] words, “A bridge is a (strong or weak) relationship for which there is no effective indirect connection through third parties. In other words, a bridge

is a relationship that spans a structural hole.” A lack of relationships among social actors, or groups of actors, in a network gives those positioned in structural holes strategic benefits, such as control, access to novel information, and resource brokerage. Actors that fill structural holes are viewed as attractive relationship partners precisely because of their structural position and related advantages [1]. These actors are called brokers (as they fill a brokerage position) or bridges. These actors form non-redundant, often weak ties among otherwise less connected actors [3]. To use a contemporary example, Indira may have a group of Twitter college friends, with whom she has strong relationships. She may also interact via Twitter with Gerhard, whom she briefly met during an internship abroad. For Indira's friends, her weak relationship with Gerhard may be an important network path to a group of professionals abroad. It is also a non-redundant relationship, as others in his strong and immediate social networks, which may be highly interconnected with one another, are not connected to this group of professionals. Indira, then, is located in a powerful structural position in her networks as a broker or a bridge. This example demonstrates that weak ties often provide less redundant connections – or, in other words, relationships with people one's friends are not connected to. Such non-redundant weak connections are beneficial to individuals, as they gain information not available through their other social ties, such as solutions for problems, employment opportunities, etc. These ties are also advantageous because one's friends depend on that individual for this type of novel information [3]. Those with high betweenness centrality often connect users found in different groups (i.e., network clusters), which helps explain information flow across groups. Look at your data. Sort the betweenness centrality column by size (large to small). • Identify the users with the highest betweenness centrality. Who are they? what can you learn about them from the data in NodeXL? What can you learn about them by searching the web? Twitter is a directed network and therefore the flow of information or influence, for instance via a bridge can be one-way to either direction or two-way, depending on the direction of relationships with others in the network. Understanding the role of users with high betweenness, then, depends not only on the connectedness, but also on the directionality of connections. Literature about structural holes, however, has traditionally assumed full symmetry of relationships in a network [4] and the betweenness centrality metrics does not take into consideration the direction of links. When evaluating key users as bridges, you will learn much by examining their in and out degree values.

III.  Social media network analysis case studies



11.5  Network analysis

Examining the interaction between organizations and the public on Twitter, users with high betweenness centrality play a key role in reaching out to users that do not interact directly with the organization. In their research, Himelboim et  al. [5] define social mediators as “the entities which mediate the relations between an organization and its publics through social media,” as they regard mediated public relations as “communicative relationships and interactions with key social mediators which influence the relationship between an organization and its publics.” They defined social mediators as users with both high betweenness and high in-degree centrality values. • Who are the social mediators in your network? Find users with high betweenness and high in-degree centrality values. User reciprocity A relationship between two users is reciprocal, or mutual, if each user has initiated a tie with the other user (see Chapter  6). Reciprocal relationships between individuals may indicate a wide range of social attributes, such as cooperation, trust, exchange of opinions, and power balance. At the user level, the Reciprocated Vertex Pair Ratio is measured as the number of users one is connected with (alters) that are reciprocal over the total number of alters. In other words, the portion of reciprocated relationships of the total number of relationships a vertex has with others in the network. On Twitter, for example, a reciprocal or mutual relationship between two users can be established if they follow one another. If Aviva is connected with 10 other users on Twitter (whether following or being followed), and 5 of these users relationships are mutual (i.e., Aviva follows 5 users who also follow her), Aviva's reciprocity value will be 0.5. Reciprocity can be used to evaluate users' relationship building. When establishing a social media presence, users often aim to attract the attention of influential users by giving them attention (retweeting, posting hyperlinks, tagging, mentioning, etc.). Reciprocity metrics can be used to evaluate the success of this strategy. Reciprocity can also be measured for the entire network, or clusters in it, as will be discussed later in this chapter. In your dataset, sort the Reciprocated Vertex Pair Ratio column from largest to smallest.

169

typically an overall negative correlation between between users' degree (in or out) and their reciprocity values. The more connected users are, the lower their reciprocity value is likely to be. It is therefore helpful to first find the top users in your network, and then compare and contrast their reciprocity ratios. Looking at reciprocity ratios, you are probably asking yourself: are these values high? Are they low? How can you tell? The short answer is that it depends on the network and comparing them to others in the network is the most useful way to determine what is high or low. An example from a Delta Airlines Twitter topic-­network helps illustrate how to evaluate reciprocity values (Figure 11.8). Two users have about the same in-­degree values: @Itoddwood and @deptapoints. However, their reciprocity levels are quite different. @Itoddwood has a low reciprocity of 0.023, while @deltapoints has a higher value of 0.256. Comparing the two, enables you to evaluate the reciprocity metrics. One user, is reaching out to many other users, by mentioning and replying to them, resulting in a high out-degree. In this particular network, @Itoddwood has little success in having these relationships reciprocated, as its low reciprocity value indicates. In contrast, @deltapoints was much more successful in attracting reciprocated relationships. About one of four ties it has with other users, is reciprocated. Reciprocity, then, should be evaluated in comparison with other users. • In your dataset, among the top users by in-degree, find the users that stand out in terms of their relatively high Reciprocated Vertex Pair Ratio values? Who are these users?

11.5.2  Network-level metrics Overall metrics Taking a social networks approach to data analysis shifts the focus from individual characteristics of users, to their connectivity-related characteristics. You've already seen how this approach is applied to user-level metrics. Another unique characteristic of social network analysis is the focus on metrics that describe a group of users in a connected component. On the Edges worksheet, an edge is the unit of analysis. On the Vertices worksheet,

• What is the highest reciprocity ratio? You probably found one of more users who has a reciprocity value of 1. A closer look will reveal that users with such a perfect reciprocity value, are not highly connected in the network. Find their in and out degree metrics. Pretty low, right? The reason is simple: the more connected users are in the network, the less likely they are to have all their connections reciprocated. There is

FIGURE  11.8  Comparison of two Twitter users with similar in-­ degree, but very different out-degree and Reciprocated Vertex Pair Ratio values.

III.  Social media network analysis case studies

170

11.  Twitter: Information flows, influencers, and organic communities

the vertex (the user) is the unit of analysis. In the Overall Metrics worksheet, the entire network is the unit of analysis. Metrics on this worksheet will describe the network as whole. You may be familiar with these metrics from Chapters  3 and 6. Key network metrics are reviewed here, within the context of Twitter topic-networks. • Vertices. The number of users in the Twitter search network. • Unique Edges. The number of ties in the networks, excluding duplicates. For example, if @Joelle mentioned @Muhammad in two tweets within this network it will be counted as a single unique edge between @Joelle and @Muhammad. • Edges With Duplicates. The number of duplicate relationships between users. For example, if @Joelle mentioned @Muhammad in two tweets within this network, this will be considered an edge with duplicates. • Total Edges. The sum of Unique Edges and Edges With Duplicates. • Self Loops. In its original use, an edge is counted as a self loop when a user initiates a tie with itself. For instance, a user may mention itself in a tweet. However, in Twitter networks, a self loop captures all tweets that did not have a relationship embedded in them. As you recall, the Tweet relationship is used in the Edges worksheet, when there are no mentions or replies in a tweet, and it is not a retweet. The self loops metrics, then, captures the number of tweets with no relationships in the network. • Connected Components. A component is a unit of one or more users that have connections among them. The Connected Components is a simple count of these components. Note: Do not confuse components with clusters. Cluster, as discussed later, are sub-groups of users that are loosely connected to one another. Components are disconnected from other components. • Single-Vertex Connected Components. These are isolated users who are talking about the topic of your network, but in a given dataset, are not connected to others by an edge. In Twitter networks, these will be individuals who post a tweet that is not a mention, reply to, or retweet. Graph density Twitter networks vary in terms of their interconnectedness. Some networks are more tightly interconnected, by mentioning and replying to one another. In other networks, users are only sparsely connected, rarely mentioning or replying to others. Network density is measured as the number of possible or potential connections (i.e., edges), over the number of actual connections (see Chapter 6). Density values range between zero and one, and can be thought of as the percent of all possible edges that are realized. The calculation is a slightly

­ ifferent for directed and undirected networks, as did rected networks have twice as many possible edge (i.e., from vertex A to vertex B, and from vertex B to vertex A). The extent to which a network is densely interconnected impacts the rate of information flow within it. Interaction between individuals leads to shared knowledge, and shared knowledge leads to even more interaction. Granovetter [3] states that tightly interconnected individuals are typically connected by strong and redundant ties. Burt [2] notes that networks in which people are very highly interconnected are better at transmitting information. Others also demonstrate an important outcome of strongly embedded close relationships as an increase in trust between individuals, which can lead to increased information transfer. On Twitter the rate at which information is spread through a network was found to depend on its density; the greater the density, the faster information spreads [6]. • What is the density of your network? Your network size often affects whether the density value is high or low. If you have a large network, and chances are you do, the density value will be rather low. As a network grows, its density is likely to shrink. Similar to the discussion about reciprocity earlier, density should be evaluated in comparison with other networks. Within a network, it is hard to determine whether a density value should be considered low or high. Often, analysts collect several datasets of the same topic-­networks over time, which allows them to evaluate a change in density. Later in this chapter, networks clusters will be discussed. As a network may contain several clusters, this will give us an opportunity to compare clusters, in terms of their density. Graph reciprocity Reciprocity metrics at the user level were discussed earlier. At the network level, reciprocity measures the extent to which ties among a group of vertices are mutual. Reciprocity is measured as a proportion of mutual edges to the overall number of edges in a network. Values range between 0 (i.e., no mutual ties in the network) to 1 (i.e., all edge are mutual in the network). Other approaches to reciprocity metrics exist. Similarly to network density, reciprocity is associated with network size (number of vertices). The larger the network, the smaller the reciprocity are likely to be. Comparing networks in terms of their reciprocity, you should take into consideration the network size. • What is the graph density of your network?

11.5.3 Groups In social networks, smaller sub-groups of densely interconnected users – clusters – often arise (see Chapter 7).

III.  Social media network analysis case studies



11.5  Network analysis

Clusters, also referred to as communities, refer to subgroups in a network in which vertices are substantially more connected to one another than to vertices outside that subgroup. In Twitter political networks, users' exposure to tweets derives from the users they follow. Twitter users, then, are more likely to read content posted by their cluster-mates than by users in other clusters. Likewise, users in a given cluster choose to expose themselves to the same set of hubs, which serve as popular information sources among these users. Clusters, then, define the boundaries of information flow, attention and influence, among users on Twitter. Researchers have deployed an array of metaphors to describe the idea of subgroups of individuals who selectively expose themselves to politically like-minded others and their information sources. A few examples include “enclave”, “filter bubbles”, “gated communities”, “sphericules”, “monadic clusters”, and “cyber-balkans”. Sunstein's [7] term “enclave,” is also a helpful metaphor here as it describes the result of filtering news selection based on political opinions. Clusters are identified by applying a mathematical algorithm that assigns vertices (i.e., users) to subgroups of relatively more connected groups of vertices in the network. The Clauset-Newman-Moore algorithm [8], used in NodeXL, enables you to analyze large network datasets to efficiently find subgroups. As explained in Chapter 7, you can also use two other algorithms are in NodeXL: the Girvan-Newman and Wakita-Tsurumi. Identify clusters by following these steps: • In the Analysis section of the NodeXL Ribbon, click on Groups and select Group by Cluster… (Figure 11.9) • In the pop up menu, select Clauset-NewmanMoore and check the Put all neighborless vertices into one group checkbox. The latter will create a “fake” cluster to accommodate all users who are not connected to others (isolates). Click OK. (Figure 11.10) • A new worksheet will appear, called Groups. • To calculate group-level metrics: Click on Graph Metrics, then Deselect All, and then select Overall Graph Metrics and Group Metrics. Click Calculate Metrics. Note: the Groups in this worksheet are network clusters. While NodeXL uses the more generic name, groups, clusters are the specific type of groups that are calculated. The new Groups worksheet (Figure 11.11) uses groups and clusters interchangeably. The unit of analysis in this worksheet is a cluster, meaning that each row represents a cluster, and metrics describe characteristics of clusters. NodeXL assign group numbers (e.g., G1, G2, etc.) based on the cluster size (i.e., number of vertices). Find the group-level metrics. You should be familiar with all of

171

FIGURE  11.9  Choosing Group by Cluster… from the Groups drop-down menu.

FIGURE  11.10  Choosing the Clauset-Newman-Moore clustering algorithm in the Group by Cluster dialog.

them by now, as they are available at the Overall Metrics worksheet. One of your groups is the group of neighborless users, defined at the previous step. This group shows Not Applicable for some group metrics, because these are disconnected users, and therefore cannot produce meaningful network metrics. Looking at the group size (i.e., number of vertices) and volume (i.e., number of edges) you will soon find that most groups are very small. In most cases, you will

III.  Social media network analysis case studies

172

11.  Twitter: Information flows, influencers, and organic communities

FIGURE 11.11  The Groups worksheet with calculated metrics.

be interested in the largest groups, which host most users. You can use several techniques to select the group inclusion threshold. For this exercise, a simple one is suggested: look for a drop in the number of vertices in groups. In Figure  11.11, the largest group (G1) has 465 vertices, followed by G2 with 258 Vertices and G3 with 117 Vertices. The isolates group (G4), as indicated by the Not Applicable cells, has 78 vertices. The rest are much smaller and can be filtered out by using the Visibility column on the Group worksheet (see Chapter 7). • Find the largest clusters in your topic-network. • Filter out groups that are too small or not important. Earlier you learned about of density and reciprocity (and other Network-Level metrics). You should evaluate these metrics in comparison with other networks. The availability of these metrics at the cluster level helps you assess and compare groups. Group G2 is larger than G3 and yet the density for G2 (0.044) is slightly higher than G3 (0.041). As larger groups tend to show smaller density values, this finding is meaningful, suggesting that users in G2 are more interconnected than users in G3. You also learn that relationships (i.e., edges) in G2 are overall more reciprocal (0.317) than in G3 (0.291). Here again we would expect the opposite direction, which makes these findings meaningful. Look back at your data, in the Groups worksheet: • Which group is the isolates group? What portion of all users are isolated? • Find the largest clusters. These are the groups that you will further analyze. • Taking into consideration the size of the clusters, what can you tell about the density and reciprocity of these clusters? When identifying groups, two other worksheets are created: Group Vertices and Group Edges. The Group Vertices worksheet lists users by group. This is an easy

way to find which users are in each cluster. You can use this worksheet to look for commonality among users within the same cluster. As you recall, metrics in the Groups worksheet measure the interconnectivity within each cluster. The Group Edges worksheet shows the connectivity across clusters. Some clusters are more interconnected, while others are more loosely connected. Since edge are directed, you can find the frequency of edges across clusters in each direction. For example: 15 edges went from G1 to G2, indicating that in this data, 15 tweets were made by users in G1 that mentioned, retweeted, or replied to users in G2. 10 edges went from G2 to G1, indicating that in your data, 10 tweets were made by users in G2 that mentioned, retweeted, or replied to users in G1. • Examine the connectivity across your major clusters. Which pairs of clusters are more strongly interconnected? In which direction? Which pairs of clusters are less interconnected? Remember from Chapter 7 that when calculating group metrics, NodeXL automatically assigns colors and shapes to vertices, based on their cluster. Visualizing the graph, it is the default option that all users in a single group will share the same color and shape. This can be changed using the Group Options dialog found in the Groups drop-down menu in the NodeXL Ribbon. For now, choose Skip groups—don't show them in the Graph Pane. In the next section, you will learn to change shape and colors of individual users, regardless of their group affiliation. The first step is to change this default option.

11.6 Visualization The network analysis metrics explored in this chapter provide insights regarding the connectivity of your topic-network (network-level metrics), users in key positions (vertex-level metrics), and communities (clusters and their metrics). Visualizing each topic network can

III.  Social media network analysis case studies



11.6 Visualization

173

FIGURE 11.12  The Graph Pane (Raw Data).

help tell their unique story. Start visualizing your network by clicking on the Show Graph button (at the top of the graph pane). In its raw form, it will probably make very little sense (Figure 11.12), so you will need to customize the graph's visual properties.

11.6.1  User-level visual properties By now, you have already identified your top users. In particular, you have calculated users' in-degree and betweenness centrality values. Use the Autofill Columns tool to associate users' visual properties with centrality metrics (see Chapter 5). • Open the Autofill Columns dialog using the NodeXL Ribbon • In the Autofill Columns dialog, choose InDegree for Vertex Size. See Figure 11.13. • If needed, you can define the range of values that vertex size may take. To do so, click on the arrow next to the relevant visual property, selecting Vertex Size Options. See Figure 11.14. • Check the Ignore outliers box if there are outliers in your dataset (e.g., users with an extremely high In-Degree). This will ignore users that have high and disproportionate values. Such extreme values may skew the distribution of size values across vertices, making differences between other vertices indistinguishable. • Consider checking the Use a logarithmic mapping box. This will place more visual distinction between the small variations in the low values (which often include the majority of vertices) and less visual distinction between the wide variations

FIGURE 11.13  Autofill Pane: Vertices' Visual Properties.

in the high values (which often include a small minority of the vertices). • Click Autofill. Then click Refresh Graph in the Graph Pane. You can see that a few vertices are much larger than others. These are your local topic hubs. • You can change the scale in the Graph Pane in order to find the best visibility of your graph. Rather than simply identifying the key actors by their size, add labels to the key users. Because you are interested in changing users' visual properties, turn to the Vertices worksheet. • Find the In-Degree column and sort it by size, from large to small. • Find the Shape column and set it to the type Label (as described in Chapter 5) for the users with the highest in-degree. • When selecting Label as the shape, NodeXL looks for the actual label in the corresponding cell in the Label column. This cell is currently empty. Type a label in it. A good label would be either the name of the user (e.g., NodeXL Project) it its Twitter handle (e.g., @nodexl). • Refresh the graph. Can you see the labels? • Change the size manually to something larger if needed. Refresh the graph. • Change the color. Just type in a major color in the corresponding cell in the Color column (e.g. Red). Refresh the graph.

III.  Social media network analysis case studies

174

11.  Twitter: Information flows, influencers, and organic communities

FIGURE 11.14  Autofill Pane: Vertex Size Options.

• Continue and assign labels and colors to all your top in-degree users. Be creative. You can select labels for all top in-degree users and select different colors, based on the identity of these users (e.g., news media, bloggers, politicians, etc.). • You can also make the changes, by right-clicking on a vertex in the graph and selecting Edit selected vertex properties. • You may notice than other nodes may cover the labeled nodes. To resolve this issue use the Autofill Columns dialog to set the Vertex Layout Order to In-Degree.

11.6.2  Cluster-level layout and visual properties NodeXL provides an array of network layout options. It is often helpful to layout the network by groups, so clusters are highlighted. Earlier, you got NodeXL to disregard groups in the graph pane. Now, though, reversing this will let you better organize your graph layout: • Open the Group Options from the Groups dropdown menu on the NodeXL Ribbon. • Deselect Skip Groups. • You can decide where the visual properties (Shape and Color) for vertices come from, the Vertices or the Groups worksheets. • Keep the Colors coming from the Groups worksheet (the first option). • Select the Vertices worksheet as the source of Shapes (the second option). This will allow you to keep the hub labeling. Next, layout the graph by clusters: • In the Graph Pane click on the layout algorithm drop down menu (see Chapter 4). Choose Layout Options as shown in Figure 11.15.

FIGURE 11.15  Layout Options.

• In the Layout style, select the second option: Lay out each of the graph's groups in its own box. Remember the clusters that you identified earlier in this chapter (Section 11.5.3)? You can now use NodeXL to highlight them, by visually separating them in boxes.

III.  Social media network analysis case studies



175

11.7  Analysis of content

FIGURE 11.16  Graph Layout by Groups.

• Box layout algorithm has several options. Select Treemap. • Click OK. Your graph should resemble Figure 11.16. • Change to a different algorithm in the Box layout algorithm. Choose the one you like the best. Laying out the network by groups, also allows you to customize each group (see Chapter  7). You will find the visual properties for groups in the Groups worksheet. Adding a label to a group can help describe what the users in the group have in common. • • • •

Change the color of G1. Add a label to G1. Label the Isolates cluster as such. Refresh the graph.

This chapter showed you how to identify users with the highest in-degree values. As clusters define the boundaries of information flow among users, you now know that not all users will be exposed to the same list of highly connected users. Now find the top users by cluster. • In the Vertices worksheet, make sure that rows are sorted by in-degree, highest to lowest. • In the Groups worksheet, click on your largest group (not isolates). • Can you see it highlighted in red in the graph pane? • Select the Vertices worksheet. You will find the vertices in that cluster highlighted as well. Scroll up and find

out the top in-degree users in that cluster. Looking at the network's highest in-degree users, the highlighted ones are associated with the selected group. • Repeat for each of the top clusters. • Some of the top users already appear as labels in the graph, from an earlier exercise. Make sure that the top users for each of the top cluster have their Shape set to Label. Explore your clusters. Who are the key users in each of the major clusters? In this example, one may conclude that the top two clusters (G1, G2) are left of the political center, while G3 is more to the right. Note that the clusters G1 and G2 are more connected, while G3 is less connected to the rest of the network. You can explore the Group Edges worksheet to quantify cross-cluster connections. What can you conclude about the interactions across the political spectrum? We will continue and explore the unique characteristics of clusters by looking into the content of the tweets.

11.7  Analysis of content The metrics and analyses discussed so far in this chapter have focused on network connectivity among users. NodeXL also calculates content-related metrics (see Chapters  6 and 8). As you recall, the Edges worksheet provides information about tweets, including the tweet itself, hyperlinks and hashtags in it. This content

III.  Social media network analysis case studies

176

11.  Twitter: Information flows, influencers, and organic communities

FIGURE 11.17  Twitter search network top items metrics.

can be aggregated to describe the entire network and clusters in it. • Open the Graph Metrics dialog (Figure 11.17) and select Network top items and click Options. Make sure that this is the only one selected. • Select Tweet as Status Column. • Click Add → Column Containing the Items → URL in Tweet → OK. This identifies the top URLs. • Click Add → Column Containing the Items -> Hashtags in Tweet -> OK. This identifies the top Hashtags. • Click OK and then click Calculate Metrics. • For larger datasets, this may take a little time.

A new worksheet was just added to your NodeXL workbook called Network Top Items. Find it and click on it. This new worksheet reports frequencies of ­content and key users by type and by cluster (for the first 10 groups). Reviewing it, make sure you recall the top clusters you identified and which of the clusters is the neighborless one. Examine the values associated with these “top 10 lists,” to determine how meaningful they are for your analysis. • Top URLs. The frequency of of hyperlinks is calculated and the top 10, for the entire network and by cluster, are reported. Top URLs can trace individual hyperlinks that were shared (e.g., retweeted) many times among users in the network, or within a cluster.

III.  Social media network analysis case studies



177

11.10  Researcher's agenda

• Top Domains. The domain is the core of any hyperlink. Put simply, it refers to the website that the hyperlink is in. Top domains is an indicator for sources of information commonly referred to by users within a topical network. For example, the domain for the URL: http://nodexlgraphgallery.org/Pages/Graph. aspx?graphID=55977 is nodexlgraphgallery.org • Top Hashtags in Tweet. The most popular hashtags found in tweets. • Top Words. The most common words in tweets. • Top Word Pairs. The most common words that show up together in tweets.

an account and using it to upload your graph, is that you will be able to revise or remove your uploaded graph. • Visit nodexlgraphgallery.org and find your graph. • You can now share it. If you click on the Tweet button, you will find a suggested tweet format. This format includes, aside from a link to the graph and #nodexl, also the key users you found in the network. This allows you to attract the attention of these users. You can also include an image of the graph in your tweet, for better attention.

Metrics in the “Network Top Items” worksheet can provides a glance into the conversation. One helpful use for these metrics is identifying the unique characteristics of clusters. Examine the metrics by cluster:

11.9  Practitioner's summary

• What are the key characteristics of each cluster, in terms of URLs, hashtags and words? • Comparing the clusters, which metrics are most meaningful in distinguishing clusters from one another? In other words, which findings here are unique to clusters? • Write up, just a sentence or two for each of the top clusters, to describe each cluster. • Label each cluster, using the Groups worksheet, or refine the labels you drafted earlier. If you are using the sample dataset provided in the beginning of this chapter, you can compare your work with this analyzed network: http://nodexlgraphgallery. org/Pages/Graph.aspx?graphID=172972

11.8  Share your work on the NodeXL graph gallery Congratulations! You have analyzed and visualized your network. Now, if you collected your own data for this chapter, make it public. The NodeXL Graph Gallery is a repository of network data uploaded by NodeXL users. This is a great way to communicate your findings to others and share it via social media posts. Since Twitter search network data is public by default, it is reasonable to share publicly. • Select the Export drop-down from the NodeXL Ribbon and choose To NodeXL Graph Gallery… • Most metadata is already set. Update the Title so it reflects your data (e.g., NodeXL Twitter data collected on 10-23-2015). • Check Also includes the workbook and its options to allows other users to download and reanalyze your network. • If you do not have a NodeXL Graph Gallery account, you can login as a guest. One advantage of creating

This chapter guided you from the early stages of data collection, through the analysis of the network structure at the vertex, cluster and network-levels, and key content characteristics of tweets, to visualization of the network in a way that highlights your findings. Twitter network structures have important implications for social media practitioners. For instance, advertising, public relations and marketing professional can consider clustering analysis as a different way for segmenting their consumers. Network analysis of the Twitter networks that consumers created when discussing a brand can capture these subgroups, as well as the key information sources and distinctive content characteristics they post and share (i.e., via posted URLs). Furthermore, social network analysis allows brand managers to identify keys users in the network (e.g., by in-degree and betweenness centrality) and consumers that allow the brand to reach out to other users that do not interact with it directly. One useful way to think about clusters in brand communities is in terms of Direct Consumers that form clusters surrounding the brand account on Twitter, and Indirect Consumers, that are captured by other clusters. Network analysis allows social media practitioners to identify clusters of indirect communities and the characteristics that helps the brand reach out to them successfully.

11.10  Researcher's agenda Researchers often aim to interpret network structures to understand patterns of information flow and consumption in social media. The Selective Exposure theoretical framework suggests that individuals prefer turning to information sources that they agree with [9]. Examining the top in-degree users within each cluster can shed light on the process of selective exposure, and under which topics it is more or less likely to take place. Furthermore, users with high betweenness centrality can bridge and possibly decrease the effects of selective

III.  Social media network analysis case studies

178

11.  Twitter: Information flows, influencers, and organic communities

exposure. Homophily provides another framework to understand how people choose to interact with others. It is defined as “a basic organizing principle” that “a contact between similar people occurs at a higher rate than among dissimilar people” [10]. Network clusters capture sub-groups of people who selected to interact with one another more than with others. The Twitter search network top items provides vital information for identifying the characteristics of each cluster of users, including their similarities in content posted, key users mentioned, etc. The analysis discussed in this chapter can address questions such as: What types of users mediated the flow of information across clusters? What type of content flows across clusters? How does the characteristics of these unique users and content differ across different content-type Twitter conversations?

References [1] R.S.  Burt, Structural Holes: The Social Structure of Competition, Harvard University Press, Cambridge, MA, 1992. [2] R.S.  Burt, Brokerage and Closure: An Introduction to Social Capital, Oxford University Press, New York, NY, 2005. [3] M. Granovetter, The strength of weak ties, Am. J. Sociol. 81 (1973) 1287–1303. [4] L.C. Freeman, Centrality in social networks: conceptual clarification, Social Networks 1 (3) (1979) 215–239. [5] I.  Himelboim, G.J.  Golan, R.J.  Suto, B.B.  Moon, A social networks approach to public relations on twitter: Social mediators and mediated public relations, J. Public Relat. Res. 26 (4) (2014) 359–379. [6] K. Lerman, R. Ghosh, Information contagion: an empirical study of the spread of news on Digg and Twitter social networks, ICWSM 10 (2010) 90–97. [7] C.  Sunstein, Republic 2.0, Princeton University Press, Princeton, NJ, 2006. [8] A. Clauset, M.E.J. Newman, C. Moore, Finding community structure in very large networks, Phys. Rev. E 70 (2004) 066111.

[9] N.J. Stroud, Selective exposure theories, in: The Oxford Handbook of Political Communication, Oxford University Press, Oxford, UK, 2017. [10] M.  McPherson, L.  Smith-Lovin, J.M.  Cook, Birds of a feather: Homophily in social networks, Annu. Rev. Sociol. 27 (1) (2001) 415–444.

Suggested reading J.K.  Harris, S.  Moreland-Russell, R.G.  Tabak, L.R.  Ruhr, R.C.  Maier, Communication about childhood obesity on twitter. Am. J. Public Health 104 (7) (2014) e62–e69, https://doi.org/10.2105/ AJPH.2013.301860. M.  Choi, Y.  Sang, H.  Woo Park, Exploring political discussions by Korean twitter users: a look at opinion leadership and homophily phenomenon. Aslib Journal of Information Management 66 (6) (2014) 582–602, https://doi.org/10.1108/AJIM-11-2012-0089. I.  Himelboim, M.A.  Smith, B.  Shneiderman, Tweeting apart: Applying network analysis to detect selective exposure clusters in twitter. Commun. Methods Meas. 7 (3–4) (2013) 195–223, https://doi.org/10. 1080/19312458.2013.813922. I. Himelboim, S. McCreery, M. Smith, Birds of a feather tweet together: Integrating Network and Content Analyses to Examine CrossIdeology Exposure on Twitter. Journal of Computer-Mediated Communication 18 (2) (2013) 154–174, https://doi.org/10.1111/ jcc4.12001. Y.T. Alfarhoud, The use of twitter as a tool to predict opinion leaders that influence public opinion: Case study of the 2016 united state presidential election, World Scientific Book Chapters (2017) 191–206. H.G.  Ramírez, R.M.G.  García, Social networks of teachers in twitter, in: Information Management and Big Data, Springer, Cham, 2015, pp. 133–145. Hansen, D., Smith, M., Shneiderman, B., EventGraphs: Charting collections of conference connections. Hawaii international conference on system sciences. Forty-forth annual Hawaii international conference on system sciences (HICSS). January 4–7, 2011. Kauai, Hawaii. M.  Smith, L.  Rainie, B.  Shneiderman, I.  Himelboim, Mapping twitter topic networks: From polarized crowds to community clusters, in: Pew Internet and American Life, 2014. Available: http:// www.pewinternet.org/2014/02/20/mapping-twitter-topicnetworks-from-polarized-crowds-to-community-clusters/.

III.  Social media network analysis case studies

C H A P T E R

12 Facebook: Public pages and inter-organizational networks O U T L I N E 12.1 Introduction to Facebook: The social graph of 2 billion people

179

12.2 Facebook networks 12.2.1 Facebook API limits

180 180

12.3 Organizational networks: Fan pages 12.3.1 Preparing the data collection 12.3.2 Data collection 12.3.3 Getting to know the data

180 180 180 181

12.1  Introduction to Facebook: The social graph of 2 billion people Facebook is a vast set of social and communication services that form a variety of interlocking networks. Facebook is an example of a “publicly articulated network” [1]. Unlike private closeness networks, these are networks we consciously curate and publicly show to others. In social media like Facebook, Twitter, and YouTube, publicly articulated networks are also used as a means for managing access to information. In Facebook users identify others as “friends” and content from these friends is potentially displayed in the interface in a rolling list of information called a news feed. Facebook has expanded from its core social network functions to include a wide range of modes of interaction via games, chats, private messages, photo, video and live stream sharing, and event planning. The Facebook friend list works as a launching pad for a host of activities that people want to do online with some people, but not everyone. Facebook is the world's largest publicly articulated social network with over 2.3 billion members as of this writing. Amazingly, over 1.5 billion of these users log into Facebook every day.1

1

12.3.4 Network analysis 12.3.5 Visualization

182 184

12.4 Practitioner’s summary

185

12.5 Researcher’s agenda

185

References

186

Suggested reading

186

The following is both an extended discussion on how to analyze and interpret like-based networks of Facebook fan pages and an additional case study of network visualization that extends much of the earlier discussions in Part II (Chapters 4 through 8). As seen in other chapters in this volume, there are many ways to create a network from the data generated by social media interactions. Facebook networks can be created from a list of who messages whom, who is featured in the same photo, who is a part of the same group, and many more explicit and implicit connections. A growing body of literature uses data from these various types of connections, as well as content that is posted, to predict the strength of ties between two people (e.g., [2–3]). Facebook once was largely open and data on the platform was easily accessed. Over the past several years, access has been limited and has now become constrained to data about fan pages. Several of the remaining network datasets historically available via the Facebook API, such as ones related to Facebook Groups and Profile Pages, are no longer available.

https://zephoria.com/top-15-valuable-facebook-statistics/.

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00012-1

179

© 2020 Elsevier Inc. All rights reserved.

180

12.  Facebook: Public pages and inter-organizational networks

12.2  Facebook networks Users interact with content on Facebook in a variety of ways, including a range of reactions to posts, posting comments, and sharing posts with their Facebook friends. Most of these interactions are private. They cannot be observed by users outside one’s set of friends and typically cannot be captured by third party tools. In contrast, the networks that form when users engage with content within publicly available pages (fan pages) are visible. Engaging with content on a public Facebook page, users form networks of shared interests when they react, comment and share content. There are several ways to look at Facebook network data. In this chapter you will focus on analyzing the connections between Facebook fan pages. This is a good example of how social network analysis can focus on the connections between organizations, as opposed to the connections between individuals.

12.2.1  Facebook API limits Facebook has limited the data it provides publicly for free, but some data is provided on relatively liberal terms. Facebook fan page data access is limited in terms of the speed data is delivered but no time frames are excluded, in contrast to platforms like Twitter that limit public access to the past 7 or 8 days at any time. Facebook provides as much historical data from fan pages as requested although the download may take quite a while, since the download is stopped and started to stay within the query per time budget allowed. Please note that when downloading network data from very popular fan pages, you may end up with very large excel files (100 MB and more) that might be too large to manage on smaller computers. Such large files can easily max out your machine's system memory when performing SNA metrics calculations. The solution to this to carefully configure your Facebook network data importer.

12.3  Organizational networks: Fan pages 12.3.1  Preparing the data collection Facebook fan pages are a popular way for a group, company or organization to present itself. The fan page network captures inter-organizational connectivity on Facebook. As such, the starting point must be the identification of a list of fan pages of interest. Examples of lists could include all public Facebook pages related to a given organization (e.g., all Facebook pages for the CDC: Centers for Disease Control and Prevention, the United Nations, the U.S. State Department, or other agencies),

pages of organizations associated with a given cause (e.g., LGBT rights or vaccination promotion), news media pages, pages associated with a political race, or pages of corporate competitors in the same market. Step one is to locate the public fan pages of these organizations. Accuracy and attention to details are critical. The name of each Facebook page can be found in the URL for each page right after the Facebook domain name (e.g., for the CDC, their page URL is www.facebook.com/CDC, so the page name for your query is the complete URL or simply CDC.). When you search Facebook for fan pages, only use results that appear under the Pages tab. Using results from other tabs will cause an error when you try to collect data.

12.3.2  Data collection NodeXL includes a built-in importer for Facebook Fan Pages. The importer does not require that you are an administrator of any fan page. The following steps describe how to use the import dialog shown in Figure 12.1. ⦁ Select the Import drop-down menu from the NodeXL Ribbon and choose From Facebook Fan Pages Network. Make sure you do not choose the From Facebook Fan Page Network (singular). That will open a different importer that can only be used if you are an administrator of a Facebook fan page. ⦁ In the Import from Facebook Fan Pages Network dialog (Figure 12.1) type all the page names (or URLs) that you identified. Separate pages by a space or a comma. We will call each of these pages a Seed page. Facebook data collection can take a long time, so at least for this exercise, limit your search queries to fewer than a dozen pages (Figure 12.1). ⦁ Set the Levels to include to 1.5. This will include a complete network of each Seed Facebook fan page, other pages each Seed page liked or were liked by (i.e., alters), connections among these alters, and any connections with other selected Seed pages or their alters. ⦁ Ideally, you do not want to Limit number of likes, unless a page has a very large number of alters. However, for quick data collection or testing purposes, you may want to limit the number of likes to 100 by clicking the appropriate checkbox. ⦁ Unselect the checkbox next to Import only the fan pages I am interested in. Checking this would only consider links between the Seed fan pages. In most cases, only including the Seed fan pages is too restrictive, so you will want to uncheck this box. ⦁ Click Login. If it is the first time, you will need to login to the Facebook page that will be opened.

III.  Social media network analysis case studies



12.3  Organizational networks: Fan pages

181

FIGURE 12.1  Data importer for Facebook fan pages including Seed fan pages separated by commas and the option chosen to collect a 1.5 level network.

⦁ Click Download. ⦁ Data collection may take time. Even an hour or more, depending on the number and popularity of the pages. You are encouraged to collected your own data. If you like, you can also use one of the two sample datasets that we provide: ⦁ A sample dataset of pages for organizations both for and against gun control can be downloaded from the NodeXL Graph Gallery at https:// nodexlgraphgallery.org/Pages/Graph. aspx?graphID=151921 ⦁ A sample dataset of all CDC pages can be downloaded from the NodeXL Graph Gallery at https://nodexlgraphgallery.org/Pages/Graph. aspx?graphID=150196 Graphs, images and examples provides in this chapter are from the Gun Control dataset.

12.3.3  Getting to know the data The NodeXL importer retrieved the like-based relationships among the fan pages, as well as some page-­ specific information. The Edges Worksheet Select the Edges tab in the NodeXL file, where a pair of connected fan pages is the unit of analysis. You will find the list of like-based relationships, with the fan page in Vertex 1 liking the fan page in Vertex 2 (Figure  12.2). Since only 1.5 levels of edges were selected for inclusion in the network, the edge list includes like-relationships between seed fan pages and their alters (Network Level equal to One) and between alters (Network Level equal to OnePointFive). As you analyze and visualize the network, all relationship-based metrics will be found on this worksheet.

III.  Social media network analysis case studies

182

12.  Facebook: Public pages and inter-organizational networks

FIGURE 12.2  The Edges worksheet showing the automatically imported Relationship, Type, and Network Level columns.

The Vertex Worksheet Let us now turn to the Vertices worksheet where a fan page (i.e., the Vertex column) is the unit of analysis. Some useful information includes a link to each fan page of Facebook, a direct link to the profile image that can be utilized for visualization (i.e., the Picture column), and a self description of the page for easy interpretation (i.e., the About column). As you analyze and visualize the network, all fan page-based metrics will be found here. Take a moment to become familiar with the fan pages in the dataset (Figure 12.3).

12.3.4  Network analysis Think about the network as a topic or an issue-based community of organizations. The selection of the pages—e.g., a political race, a social cause or an industry—determines the domain of interest, interaction, and informing flow. Like Twitter-topic networks (see Chapter  11), all social network metrics must be evaluated within the boundaries of the network. Vertex-level analysis Ties between fan pages are directed, since “liking” a page is a unidirectional action. This is unlike friendship-ties among two Facebook users’ personal ­

pages, since “friending” only occurs when there is a mutual agreement. Use the Graph Metrics tool (see Chapter 6) to calculate the In-Degree, Out-Degree, Betweenness Centrality, Reciprocity, PageRank, and Overall Metrics. In-degree centrality measures the number of fan pages that “liked” a given fan page. It captures attention given to an organization or another public entity on Facebook. A community of users who are interested in a given topic or an issue are more likely to become aware of content posted by a page with a high in-degree centrality, either as it appears on those pages that “liked” it, or by following links to the page. In contrast, out-degree centrality measures the number of fan pages a given page “liked.” This is an indication of the attention a fan page gives to other pages, and the channel that this page opens to its visitors to other related pages. Sort on the In-Degree and Out-Degree columns and examine which fan pages fill these important network positions. Betweenness centrality helps identify bridge spanners who connect other fan pages that would not otherwise be as closely connected (see Chapter 6). Remember that this metric is calculated based on undirected ties. As a result, when examining fan pages with high betweenness centrality, you should also look at the in-degree and out-degree to determine what is driving the high scores. For example, a fan page with a high betweenness

FIGURE 12.3  The Vertex worksheet showing the automatically downloaded Facebook fan pages fields.

III.  Social media network analysis case studies



12.3  Organizational networks: Fan pages

c­ entrality and in-degree indicates that the page is “liked” by other fan pages who have not “liked” each other (i.e., fan pages who are likely less aware of each other). In contrast, a fan page with high betweenness centrality and out-degree indicates that the fan page “likes” many disparate fan pages that are otherwise not likely to have “liked” each other. Sort on the Betweenness column to identify the pages that have the highest scores. Which fan pages are the primary connectors?. PageRank centrality is another useful metric that captures important fan pages that considers the number of other fan pages that “like” the page, as well as characteristics about the linking fan pages such as their centrality and link propensity (see Chapter 6). Fan pages that rank highly are ones that are “liked” by other popular fan pages. They are the ones who those “in the know” are pointing toward. PageRank considers the directed nature of the network, unlike Eigenvector Centrality, which shares many characteristics, but is designed primarily for undirected networks. Sort on PageRank and consider who is ranked high and why. Reciprocity measures the mutuality of directed ties. In this network, the vertex-level reciprocity (i.e., Reciprocated Vertex Pair Ratio) measures the percent of cases where one fan page “liked” another fan page, which in turn “liked” the original fan page. You should be cautious not to simply find the pages with the highest reciprocity ratio, because these are likely to be the pages with the least number of likes. Instead, evaluate levels of reciprocity of nodes already identified as located in key positions in the network (e.g., High centrality metrics already discussed). Examine the key users that you already identified earlier. Which have the highest reciprocity? Which have the lowest reciprocity? Why might that be? Cluster-level analysis A cluster, or subgroup of more tightly interconnected nodes, captures the naturally occurring and self-selected boundaries of information flow within a given network (see Chapter 7). Decades of literature highlights the tendency of individuals to connect with similar others, a phenomenon known as homophily, or “Birds of a Feather Flock Together” [4]. In Facebook's fan networks, clusters emerge when groups of pages “like” one another. Within a given domain of interest, clusters can capture, often hidden, sub-groups of pages that share common interest, common causes or even reflect factions within a larger community devoted to a common cause. Calculate network clusters by choosing Group by Cluster from the Groups drop-down menu in the NodeXL Ribbon. Try out the three different clustering algorithms to see if one seems to create groups at the most meaningful level of granularity. Then make sure and use the Graph Metrics dialog to calculate Group metrics.

183

With the aim of understanding the cluster-structure of your network, examine cluster-level metrics and fan pages’ characteristics. The Groups worksheet includes metrics about each cluster, including number of vertices (fan pages), number of edges (likes) and density (level of interconnectedness). It also provides visualization opportunities that will be discussed in Section  12.3.5. Which are the major clusters in the network? How are they different in terms of size, volume, reciprocity and density? Note that cluster size is inversely correlated with density, so larger clusters tend to have lower density. As you study clusters, compare similarly-sized clusters or patterns that break the expected relationship between smaller-sized groups with lower density. The Group Vertices worksheet lists the vertices—here fan pages—by cluster. The best way to understand what makes each cluster unique is to understand the similarities among fan pages within the same cluster, and their differences from fan pages in other clusters. Examine fan pages by cluster to understand each one's unique characteristics. As you may discover, finding commonalities among a large number of fan pages can often be cumbersome. Utilizing vertex-level metrics can help identifying fanpages in unique positions within each cluster. A fan-page with high in-degree within a cluster suggests that many other fan pages liked it. A fan-page with high out-degree within a cluster would suggest that the page initiates ties with many other fan pages. The top in- and out-degree users by cluster can provide insights toward understanding the unique characteristics of each cluster. Many of the high degree users will be Seed fan-pages that were used to collect the initial data. Finding which Seed-pages are clustered together and which non-seed pages are central in the network can be helpful here. Take the next simple steps to identify users with high centrality by cluster. ⦁ In the Groups worksheet, identify the top clusters. It can be all of them, or you may choose to disregard very small clusters by setting their Visibility to Skip. ⦁ In the Vertices worksheet scroll all the way to the right to find the Vertex Group column. ⦁ Use the Sort A to Z, Text Filters features (available from the drop-down menu in the column header) to select one group cluster at the time. ⦁ Organize rows by in-degree centrality to find the top for each cluster. ⦁ Organize rows by out-degree centrality to find the top for each cluster. ⦁ Repeat this process for additional Vertex Groups From a network structure standpoint, clusters may vary in terms of their distinctness. The more separate the clusters, the more meaningful it is to analyze the network in terms of its emerging groups. The modularity metrics, found in the Overall Metrics worksheet provides this

III.  Social media network analysis case studies

184

12.  Facebook: Public pages and inter-organizational networks

i­mportant measurement. Modularity ranges between zero, suggesting completely overlapping clusters, and one, completely separated clusters. As a rule of thumb, values above .6 are indicative of high level of modularity. Values between .4 and .6 are a moderate level of modularity, and values under .4 are a low level of modularity. Higher modularity signals that the unique characteristics of clusters are important. You now have a lot of information about each cluster. Aggregate it and carefully characterize the unique characteristics of each important cluster. ⦁ Calculate the Group metrics to find commonalities within clusters and differences across clusters. ⦁ Which seed fan pages are clustered together? ⦁ Which seed organizations better connect clusters (i.e., have high betweenness centrality)? ⦁ Which non-seed organizations capture unique positions in the network

12.3.5 Visualization At this point of the analysis you should have a good grasp of the key users in the network and what makes each cluster unique. This section shows you how to visualize fan page networks (Figure 12.4). Click Show Graph. The results may not initially look very

­ eaningful. However, if the network is similar to the m Gun Control network shown in Figure 12.4, you may see some clear clusters that could be visually separated to add clarity to the visual. In fact, it is not even clear from the current visualization that several of the clusters are not connected to each other at all (i.e., they are distinct connected components). In order to layout the graph by its key clusters, choose Layout Options... from the layout drop-down menu and choose the Lay out each of the graph's groups in its own box option (see Section 7.3.7 for instructions). Now, use your knowledge from earlier chapters to visualize the network in a way that tells its story (Figure 12.5). This may include using the Size and Label options. You may even want to use Color or Shape to indicate which pages are Seed pages. Remember, to change the shape of an individual vertex, you must first choose Group Options from the Groups drop-down menu in the NodeXL Ribbon and specify which worksheet the color and shape should be pulled from. Now, try and make sense of the clusters. Looking at the graph, see how the pro-gun control clusters are more heavily connected, while the pro-guns cluster is almost completely disconnected from others. Learn more about the public pages in each cluser, by examining the links to the Facebook fan pages on the Vertices worksheet.

FIGURE 12.4  Graph layout (not by cluster).

III.  Social media network analysis case studies



185

12.5  Researcher’s agenda

iraqveteran8888official

Gunowners Everytown

Pro gun 1

Randpaul

Mikebloomberg

Stophandgunviolence

Bradycampaign

Momsdemandaction

Momsdemandactionnewjersey

Coalitiontostopgunviolence

Pro gun control 2

nagrfb

Pro gun 2 Policital fugures (top) and pro gun control 1 (bottom)

News outlets

Pro gun 3

FIGURE 12.5  Gun Control network after grouping by cluster, sizing based on betweenness centrality, and labeling the highest betweenness centrality fan pages.

Last, upload the graph to NodeXL Graph Gallery: ⦁ Choose the Export drop-down in the NodeXL Ribbon and then select To NodeXL Graph Gallery ⦁ Edit the Title to describe your network ⦁ Note that your seed fan pages are already included in the description section. ⦁ You can create an account in nodexl here or upload your graph as a guest. ⦁ Visit NodeXL Graph Gallery to see your graph.

interconnected structure of its related Facebook pages, allowing Facebook users to easily navigate across a variety of brands, for instance, associated with a company. On the other hand, each Facebook page can become a gateway to different groups of audiences (i.e., network clusters), to maximize the exposure of a company’s content on Facebook. Identifying clusters of pages, including their levels of interconnectivity can help an organization find an appropriate balance between reaching out to the common audience and using its affiliated account to bridge out to new audiences.

12.4  Practitioner’s summary Social media practitioners often seek to evaluate their organizational social media activity and suggest strategies for improvement. Facebook networks can also capture intra-organizational dynamics, for organizations that are large enough to have many public pages. On the one hand, an organization can benefit from a coherent and

12.5  Researcher’s agenda Researchers are interested in examining the interactions among social entities larger than an individual, such as companies, organizations, and countries [5]. The Facebook networks that are formed based on the interconnectivity among

III.  Social media network analysis case studies

186

12.  Facebook: Public pages and inter-organizational networks

public pages can be explored using the body of literature about organizational networks. From an ­inter-organizational standpoint, Facebook networks can be considered a proxy for relationship among organizations in real life. The sample dataset of guns-related organizations makes a good illustration for the value of social network analysis. Clusters capture the expected divisions between and the two opposing movements, but also factions within each movement. Users with high betweenness centrality, especially if not part of the seed organizations, are of great importance as they have the potential of bridging factions within a movement, serving as gatekeepers of information flow. Burt’s theory of structural holes [6] can shed light on identifying locations in the network, that if filled can give great advantage to that organization, in becoming a broker, especially in sparse or divided networks.

References [1] A.  Zollers, Critical perspectives on social network sites, in: R.  Hammer, D.  Kellner (Eds.), Media/Cultural Studies: Critical Approaches, Peter Lang Publishing, Inc., New York, NY, 2009. [2] E. Gilbert, Predicting tie strength in a new medium, in: Proceedings of the ACM 2012 Conference on Computer Supported Cooperative Work, 2012, pp. 1047–1056. [3] J.J. Jones, J.E. Settle, R.M. Bond, C.J. Fariss, C. Marlow, J.H. Fowler, Inferring tie strength from online directed behavior, PloS One 8 (1) (2013) e52168.

[4] M.  McPherson, L.  Smith-Lovin, J.M.  Cook, Birds of a feather: homophily in social networks, Annu. Rev. Sociol. 27 (1) (2001) 415–444. [5] M.  Kilduff, W.  Tsai, Social networks and organizations, Sage Publications Ltd, London, England, 2003. [6] R.S. Burt. Structural holes and good ideas, Am. J. Sociol. 110 (2), 349–399.

Suggested reading J. van Loon, L. Unsöld, The Work and the Net: a critical reflection on Facebook-research methods and optical mediation, J. Curr. Issues Media Telecommun. 6 (1) (2014). B.  Rieder, Studying Facebook via data extraction: the Netvizz application, in: Proceedings of the 5th Annual ACM Web Science Conference, 2013, pp. 346–355. S. Catanese, P. De Meo, E. Ferrara, G. Fiumara, Analyzing the Facebook friendship graph, arXiv 1011 (2010) 5168. A.  Oshri, I.  Himelboim, J.A.  Kwon, T.A.  Sutton, J.  MacKillop, A.M. Kogan, Childhood physical and sexual abuse and social network patterns on social media: associations with alcohol use and problems among young adult women, J. Stud. Alcohol Drugs 76 (6) (2015) 845–851. A. Majumdar, D. Saha, P. Dasgupta, An analytical method to identify social ambassadors for a mobile service provider's brand page on facebook, in: Applications and Innovations in Mobile Computing (AIMoC), 2015, pp. 117–123. H. Wang, Cultivating a fan base on Facebook for public health promotion: the case of East Los High, in: Medicine 2.0 Conference, 2014. JMIR Publications Inc., Toronto, Canada. J. Ryu, S. Lee, M. Byun, I. Lee, Study of fashion brands’ Facebook fan pages using social network analysis, 2016.

III.  Social media network analysis case studies

C H A P T E R

13 YouTube: Exploring video networks Itai Himelboim*, Jen Golbeck†, and Bryan M. Trude‡ *

SEE Suite: Social media Engagement & Evaluation lab, Department of Advertising & Public Relations, Grady College of Journalism and Mass Communication, University of Georgia, Athens, Georgia †College of Information Studies, University of Maryland, College Park, MD, United States ‡Department of Advertising & Public Relations, Grady College of Journalism and Mass Communication, University of Georgia, Athens, Georgia O U T L I N E

13.1 Introduction

187

13.2 What is YouTube?

188

13.3 YouTube’s structure 13.3.1 Videos 13.3.2 The user channel

189 189 189

13.4 Networks in YouTube 13.4.1 Video networks 13.4.2 Users’ networks

189 189 190

13.5 Hubs, groups, and layers: What questions can social network analysis of YouTube answer? 190 13.5.1 Video networks 190 13.5.2 User networks 191 13.6 Importing YouTube data into NodeXL 13.6.1 Importing video data

191 192

13.1 Introduction Billions of videos about wildly diverse topics have been uploaded to the Internet by hundreds of millions of people. Using the techniques of social network analysis, you can visualize the landscape of connected videos and users to highlight important patterns that link the producers, commentators, and consumers. In just the past few years, online video sharing has become a growing mainstream social practice. Gone are the days when watching a video online was an onerous task, involving the installment of media players and a ­prolonged wait for the content to download. Today, people easily share private videos with friends and family; amateurs and professionals broadcast artistic endeavors, from music to comedy to directorial ­experimentation; Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00013-3

13.6.2 Importing user data 13.6.3 Ethical considerations 13.6.4 Problems with YouTube network data

192 193 193

13.7 Preparing YouTube network data

193

13.8 Analyzing YouTube networks 13.8.1 User networks 13.8.2 Video networks 13.8.3 The YouTube “makeup” video network

194 194 197 197

13.9 Practitioner’s summary

202

13.10 Researcher’s agenda

203

References

203

Suggested reading

203

media ­corporations distribute TV programs and movie excerpts, and millions of people watch and recommend videos to others, making some of them “viral” and wildly popular. Online video sharing services offer something for almost everyone, whether they are video content creators or consumers. Video content is used for many purposes: conveying knowledge, disseminating information, self-promotion, documenting world affairs, and much more. The diverse content that can be found on video-sharing sites draws large numbers of users. But not all content, even popular content, is popular in the same way. As you will see, networks of producers and consumers of online videos vary greatly. YouTube has become almost synonymous with watching videos online. Although “video sharing” is a term less familiar to most people, “YouTube” videos

187

© 2020 Elsevier Inc. All rights reserved.

188

13.  YouTube: Exploring video networks

are commonplace. Using video-sharing systems to publish or consume videos creates a variety of connections between the people who use these systems, the videos, and the tags that describe them. The social structures that emerge from the interaction of video creators and viewers can be represented as a social network graph. Participation in online video sharing generates a number of network graphs that can reveal not only users’ preferences for video content but also their habits, motivations, and social interaction. Many people upload, view, and comment on videos, but some become centers of dense communities or are well connected to others. Because of its extreme popularity, YouTube presents an especially inviting domain to explore the social structure and dynamics of networks within video-­sharing communities.

13.2  What is YouTube? Created in 2005, YouTube is an arena for personal communication, a place to create online communities or egocentric social networks, and a platform that can be used for distributing commercial content. YouTube was one of the first online services to offer users the opportunity to upload videos and share them with the world. Though similar services, such as MetaCafe, Yahoo! Video, and Google Video, emerged around the same time and were later followed by Vimeo, DropShots, and many others, YouTube has become the most popular v ­ ideo-sharing service in the world. Google bought YouTube in 2006, and it is now operating it as a subsidiary of Google Inc. Some of the site’s current features are now based on Google tools (e.g., the search and suggested list of friends based on Gmail contacts). The immense popularity of YouTube can be seen in the current statistics offered by the company: as of April 2019 there were over 1.9 billion users from 91 countries in 80 different languages who watch over 1 billion hours of video daily.1 YouTube’s popularity can be attributed to several factors: the relative ease of uploading and sharing videos, the site’s continuous design updates (which reflect the evolution of online social networking practices), and strategic collaborations with commercial content providers such as broadcast networks, movie studios, and political parties. There is an entire cottage industry that has developed to help YouTubers commercialize their content (e.g., see [1]). The foremost reason for YouTube’s success is the relative ease of uploading and sharing videos. Video sharing existed before YouTube prevailed: videos were sent as email attachments or were available through other video hosting services, yet these services were slow, 1  https://www.youtube.com/intl/en-GB/yt/about/press/.

cumbersome, and limited in the amount of storage they offered. Viewers could not watch videos instantly but had to download them to their computers before playing and viewing them using proprietary video players. Metadata descriptions about the video were rarely available. YouTube changed all that, as well as moving the viewing experience from a solitary experience to a social one. YouTube supports and encourages the embedding of videos in other forms of online communication—from email to microblogging to status updates in social networks—by displaying the relevant link address next to each video. All users have to do is cut and paste these video link addresses into other social media such as blogs, wikis, emails, and status updates. The simplicity of video sharing and embedding contributed not only to YouTube’s popularity but also to the phenomenon of “viral videos”: usually provocative, quirky, or creative videos that achieve extreme popularity after being mass-distributed through electronic word of mouth via various online social tools. The practice of video sharing creates several types of networks: some networks are based on content and others on social affinity or social ties. Content networks reflect mutual interests or shared hobbies or practices, essentially creating communities of practice [2] that stem from the commonalities between users’ interests. These communities can loosely be based on the preexisting categorical definition suggested by YouTube (e.g. “music,” “entertainment,” “how-to and style,” “politics and news,” etc.) or sub-networks of people interested in a specific aspect of the overall category (e.g., Japanese anime enthusiasts, environmentalists, cosmetic makeup aficionados, gamers). Content networks evolve around videos and are based on ties (edges) between videos (vertices), which are formed through the use of social tools for creating comments, creating linked videos, collecting “favorites,” and tagging content with keywords. Social affinity networks are formed when users interact with each other. YouTube allows users to subscribe to other users’ “channels” of video collections. Subscription ties are the edges of the networks, and users are the vertices. Social affinity networks can be based on preexisting relationships (e.g., family members, friends, or fans) or form on the site, as people interact with one another based on mutual interest in content. As you will see, the conceptual distinction between the two networks is sometimes difficult. However, as the next section explains, they are structurally separate. Analysis of different YouTube networks will allow you to reveal the key positions some people and videos occupy within the mesh of connections created when users collectively create, watch, and comment on content or form personal relationships; in some cases they may even offer insights as to the underlying reasons for these connections.

III. Social media network analysis case studies



189

13.4  Networks in YouTube

13.3  YouTube’s structure According to YouTube’s official policy, the purpose of the service is to give everyone a voice and show them the world. As such, YouTube’s structure is based on two sub-layers, clearly differentiating between videos (content) and users (community) while maintaining a close linkage between the two. Similarly, network analysis can be performed on both networks of video or users, either independently of each other or in conjunction. YouTube features are constantly changing, with new tools being introduced and others periodically removed, often directed at making YouTube an even more social space. As with almost every online social network, you should familiarize yourself with the latest changes, before analyzing a YouTube network.

13.3.1 Videos YouTube videos are displayed on separate pages with consistent layouts. YouTube pages also include controls for video playback features, a Subscribe button, a Share button, and the ability to Report inappropriate videos. Related videos are also shown, as well as metadata about the video as described below: • Title—a title chosen by the user who uploads the video. • Description—a detailed description provided by the user when first uploading the video. • Username—the poster's username and icon, which links to their channel (see Section 13.3.2). • Tags—chosen by the user to describe the video so searchers can more easily find it. These are not shown publicly on video pages, but can be viewed in the page source or using third party tools like vidIQ Vision for YouTube chrome plugin. • Category—chosen by the user from a closed-list of categories provided by YouTube. • Views, data, and statistics—YouTube provides the number of views the video generated, as well as the number of likes and dislikes, and the date of publication. YouTube Studio beta provides many additional stats for videos you have posted (e.g., location of viewers, viewer trends over time). • Comments—comments about the video by users. Comments can be threaded (see Chapter 10), pinned to the top, and liked/disliked by other users.

13.3.2  The user channel

a playlist, show a series, or highlight recent activity. They can also be reordered on the page. Thus, users’ pages can be strikingly different: some users prefer not to disclose any personal information, whereas others exhibit their social relations and ­detailed personal information publicly. Most u ­ sers display information about themselves (e.g., name, age, location, the date they joined YouTube and the last date they logged in, links to other personal websites, and a­lternative means of communication— email, instant messaging (IM), or Facebook account), although this can be limited with privacy settings. Users have the option to display two different social networks via the Channels section of the page: 1. Featured Channels. These are other YouTube channels that the current channel wants to highlight. They are often related channels or ones that the current user values highly. They create a directed (asymmetrical) edge that points from the current user’s channel to each featured channel that is listed. YouTube limits this to 100 featured channels, and many YouTubers only include a handful of them. 2. Subscriptions. These are other YouTube channels that the current channel is subscribed to. Only those that are listed as public will be listed on the Subscriptions section of the Channels page. Like Featured Channels, these are directed links that point from the current user’s channel to each user’s channel that they are subscribed to. With effort and time, a dedicated reader or researcher can piece together pieces of information that assemble an interesting picture of a user’s activity and preferences.

13.4  Networks in YouTube YouTube’s rich collection of users, videos, comments, subscriptions, featured channels, tags, ratings, and favorite videos offers multiple ways networks can be formed. Broadly, video networks are different from user networks in both content and structure. Within these networks are several subnetworks that provide insights into the important people, videos, and events in these video-sharing networks. Interesting connections between people and content can be found. Examples of these different networks are presented later in this chapter, but first you’ll need to understand the attributes of each network.

13.4.1  Video networks

Similar to other social networks, YouTube users can create personal profiles called “channels,” which are customizable. Users can choose what information to share with other viewers and which sections to display on their channel. Sections allow you to display videos (e.g., highlight popular or recent or linked videos), show

Several networks can be constructed that connect videos to other videos using the attributes found on video pages: • Videos that share the same descriptors. When users upload videos to YouTube, they must provide video content descriptions, including a title and tags or related

III. Social media network analysis case studies

190

13.  YouTube: Exploring video networks

keywords. Videos can also be classified according to predefined categories, such as comedy, music, education, politics and news, people and blogs, how-to and style, and so on. Videos that share the same descriptors may also share the same type of content; however, as users are the ones who assign video descriptors, the descriptors can vary widely and you cannot assume content similarity in videos that share descriptors. The emergent collection of connections among videos has varying degrees of density, from tightly knit networks of topical videos that discuss similar content to a dispersed network of videos that have little in common. These content network can be created using data captured in the Tags, Author, Description, and Title fields that are populated when using the Import From YouTube Video Network as described in Section 13.6.1. However, you will need to also apply techniques described in Chapter 8 to transform data in those columns into these types of networks. • Shared comment networks. Users leave textual comments about videos they have watched, often producing lively discussions. Videos can be connected to one another when the same person (or people) comment on them. For example, if Marc Smith comments on a NodeXL Intro video and a NodeXL New Feature video, an undirected edge would then connect the two videos. If Derek Hansen also comments on those two videos, the edge would have a weight of 2. This type of network can be imported using the Import From YouTube Video Network as described in Section 13.6.1. • Related videos. A list of related videos is adjacent to each selected video. These lists are based on YouTube algorithms. The current version of NodeXL does not have a way to capture these networks.

13.4.2  Users’ networks In contrast to video networks, which focus on content, user networks focus on connections between users. User networks can be explicit or implicit. The direct request or action of at least one user creates explicit networks. Users take the effort to click to create “subscription” networks and display those connections on their channel. Or, users choose to add another user’s channel to their Featured Channel list. Implicit networks are created when two or more users interact through comments, ratings, and favoriting on user and video profile pages. Of the three, only comments are visible to external users of YouTube in personally identifiable form; ratings and favoring are anonymized and summed to a single value. When one user comments on another’s channel profile, an implied connection is created between them. Not all, or even most, comments are responded to, but all create a connection and allow other users to exhibit their interest in that user and the content he or she provides.

The NodeXL Import From YouTube’s User’s Network described in Section 13.6.2 allows you to capture the network of user subscriptions. Currently, other user-to-user networks, such as a comment reply network (see Chapter  10) are not possible to capture using the importer. However, it is possible to create a user-to-user network that connects users who have commented on the same videos using data from the Shared Commenter field after using the Import from YouTube Video Network described in Section  13.6.1. This will require significant data preparation, which is not covered in this chapter. Exploring user networks can provide insights into the overall structure of video collections and key videos and users who occupy critical positions in that network, as well as contextual connections among videos and users (Figure 13.1). You will learn in this chapter about the characteristics of these networks. Understanding the nature and structure of the relationships and ties within YouTube networks can help you understand important interaction patterns and information flows within the networks.

13.5  Hubs, groups, and layers: What questions can social network analysis of YouTube answer? YouTube does not easily reveal its underlying network structure. The interface displays individual leaves and branches but not the larger forest of connections it contains. Before deciding which data to collect and analyze, you have to step back and conceptualize the questions that are of interest. Once formulated, questions may be pertinent to both video and user networks.

13.5.1  Video networks 1. Centrality. Which videos are central within a category/ type of videos? Which videos generate many comments, response videos, and higher ratings? These videos and users may influence the content produced in other videos and attract many relationships (i.e., subscriptions) with people who share an interest in that content. Some videos are central to a specific category, whereas others are peripheral. Are there differences between a single video and a series of videos produced by the same user? (Do series increase the overall popularity of individual videos? Can a single video be as pivotal as a series of videos?) 2. Groups. Does the network contain hubs of densely interconnected videos that share properties like common tags and descriptors? Which videos are central to those hubs? Is their centrality correlated to other attributes? Are different hubs connected to each other? Which are the boundary videos that connect such hubs? How dense are these hubs? How do they compare with other types of social content?

III. Social media network analysis case studies



191

13.6  Importing YouTube data into NodeXL AbsolutelyAlex The Game Theorists Em-One Smooth McGroove Atpunk

Funhaus

Brandon Turner RubberRoss tithinian

DashieGames

oddboy

Master Sword

SleepyCabin Rooster Teeth

JonTronShow LazyPillow (Yotam Perel)

DidYouKnowGaming?

Cinemassacre

GameGrumps Sunder

SuperMega

ProJared

Egoraptor

GrumpOut

Ninja Sex Party

The Game Station

RubberNinja Raleigh Ritchie

Markiplier

Mortem3r

jacksepticeye BroDuo

KittyKatGaming Commander Holly

Hot Pepper Gaming Gamer Next Door SoloTravelBlog

Give Heart Records - Nathan Sharp

FIGURE 13.1  A YouTube user's 1.5 ego subscription network.

3. Temporal comparisons. How does a video network evolve over time? What affects its development? Are certain descriptors, tags, topics, and types of videos crucial to the evolution of the network? What is the effect of rapidly and widely exchanged viral videos on the development of the video and user network? Do these videos disrupt or reinforce existing networks, or is the effect of viral videos visible mostly outside YouTube’s boundaries? What changes occur when a video becomes popular?

13.5.2  User networks 1. Centrality. Which users are central in the network of connected YouTube users? Some users may be central in a specific category but not in others. Is centrality an outcome of the explicit or implicit networks? Which users are boundary spanners between different parts of the networks? Which are peripheral? Can you identify rising YouTube stars? 2. Groups. How do users link together to form emergent groups? What brings them together? (e.g., is it a certain interest, topic, or another reason?) How do the populations of subscribers and featured channels overlap? Implicit groups can be found and compared with explicit groups. Are there central and peripheral groups? Do subscribers-of-subscribers belong to the same groups? How dense are these groups? 3. Temporal and structural comparisons. How and why does the popularity of users change? How do users

move from being peripheral to central and vice versa? Are boundary spanners changing over time? How do external circumstances affect users, their popularity, and their networks? How do the video, subscription, and friendship networks align? What are the differences between a user’s subscription and friendship network? Which is denser? Which is larger? How does a change in a video popularity affect these networks? Are there differences between the explicit networks and the implicit ones? How do they affect each other? Some of YouTube’s features, such as favoriting, number of views, and lists like “Most viewed,” “Most subscribed to,” or categories such as “Rising videos” can give you an understanding of popularity trends. But you cannot learn about information flows, centrality, and subnetwork structures from these features alone. For that, make use of NodeXL's network analysis metrics, coupled with network graph visualizations.

13.6  Importing YouTube data into NodeXL To import data from YouTube into NodeXL, first select which network you are interested in and what type of data to import using one of the data Import options in the NodeXL Data menu. When deciding what data to import, remember that YouTube differentiates between

III. Social media network analysis case studies

192

13.  YouTube: Exploring video networks

videos and users. Although networks of both users and videos can be imported and integrated into the same NodeXL network file and later compared, they cannot be imported simultaneously.

13.6.1  Importing video data From the NodeXL ribbon, choose Import then From YouTube’s Video Network. A dialog box, similar to Figure 13.2. will be displayed. • In the first field, type your search string. It can be a single word or a full Boolean string, similar to your Twitter data collection (see Chapter 11). • Next, select: Pair of videos commented by the same user (slower). Two videos then will be connected if at least one user posted comment - text of video - to both. It is helpful to think of a link as shared interest, as at least one user expressed interest in both videos by posting comments to both. • You may want to limit the number of videos, the number of top-level comments, and the number of replies in discussion threads to consider when identifying co-commenters. Use the checkboxes and numbers in the bottom three fields to do do this.

You are encouraged to collect your own data, but if you prefer, you can download the sample dataset based on the search term “eye shadow” AND “makeup” used in this chapter from here: https://nodexlgraphgallery.org/Pages/Graph.aspx?graphID=175213

13.6.2  Importing user data From the NodeXL ribbon, choose Import then From YouTube User’s Network. A dialog box, similar to Figure  13.3, will appear. This option allows you to import a network associated with YouTube users. In the search box you can indicate either a user or a channel as the Seed of your ego network: • By username. The username is displayed in the URL, right after youtube.com/user/. For instance, if the URL is www.youtube.com/user/Marc1Smith then the username to include in the importer is Marc1Smith. • By Channel. The channel ID is displayed in the URL, right after youtube.com/channel/. For example, if the URL is www.youtube.com/ channel/UCalCDSmZAYD73tqVZ4l8yJg then the channel ID to include in the importer is UCalCDSmZAYD73tqVZ4l8yJg.

Your decision about which data to import depends on your research question. Some options will substantially slow the import process and will require further filtering of the data in order to analyze it later. You may need to use the filters to set an upper limit to the number of videos that will be imported in any data import. This is very useful when considering the size of YouTube and the staggering number of videos and the limits of desktop spreadsheet programs.

You can then select the network Levels to Include. A level is a step in the network from one node to another.

FIGURE 13.2  Import from YouTUbe video network.

FIGURE 13.3  Importing user’s network.

• A single level starts with a target user and takes a single step out to all of their subscribers or subscriptions. A second level includes data about all of the friends of their friends.

III. Social media network analysis case studies



13.7  Preparing YouTube network data

• Taking a half step back, you can construct a 1.5 level network, which limits the list to the ties among first level subscribers. A tie from a friend to someone who is not the selected user’s friend will not appear while ties among their mutual friends do appear. • The higher the number of levels you choose to go into the network to collect data, the longer it will take to import the data from YouTube into NodeXL. In most cases, asking the spigot to bring back 1.5 levels of the network is sufficient to answer many common questions. You can also limit the number of users (vertices) that will be imported, if needed, to extract a dataset of manageable size. You are encouraged to collect your own data, but if you prefer, you can download the sample dataset based on the user GameGrumps used in this chapter from here: https://nodexlgraphgallery.org/Pages/Graph. aspx?graphID=175211

AD VA N C ED TOPIC Using pre-prepared data If you would prefer working with data that is not imported into NodeXL through the data imports, this data can be gathered independently and opened directly in Excel and imported into NodeXL (see Chapter 4). YouTube requires use of the YouTube API to gather data. This is maintained as part of the Google Code system and documentation is available at https://developers.google.com/youtube/v3/. The API describes how data can be accessed from YouTube. It is based in the Google Data Protocol, a REST-inspired protocol for accessing the data over the web. This means data can be accessed with precisely formatted machine-readable content over the web and through a wide variety of programming languages. You can write your own code to gather YouTube data through this API. You may want to crawl a network, beginning with a specific user, moving outward through the network. You may also want to send queries and create connections between users or videos based on criteria different from those available in the NodeXL YouTube data spigot. Once you have gathered the data, it can be loaded into Excel and formatted for use in NodeXL.

193

because of privacy preferences (for instance, complete subscription lists will not be imported into NodeXL if they are defined as private). In addition, ethical considerations stem from using other people’s data, even when publicly available. When looking at a user’s data or performing social network analysis on them, researchers should act carefully and with respect to users’ expectations of privacy. Because not all users are aware of privacy settings and related considerations, researchers should demonstrate care when dealing with information that may be perceived of as private. For example, personal ties and other personally identifying information. In the case of YouTube, there is the danger of inadvertently disclosing extremely personal, embarrassing, or sensitive information. Putting together a name, face, and an opinion may be more revealing than many users expect. Anonymizing data can alleviate some of these concerns; however, it is problematic in a network that offers many facets of extensive data about the users. Therefore the rich metadata offered along with YouTube videos should be handled with care and with respect to the users behind it.

13.6.4  Problems with YouTube network data Data collected for analysis in NodeXL, either through the importers or through your own code written with the API, is not necessarily complete or accurate. First, the YouTube data API is not 100% reliable when used by either your own code or the NodeXL importers. Because it accesses data over the web, requests may be lost or timed out. Thus, the same query may yield different results at different times. Second, even without errors, imported data may not reveal a complete network. Videos and user profiles may be marked private, preventing them from being accessed and included in the analysis. Third, users can choose to remove a video they uploaded at any time, and YouTube can remove a video when it violates the site’s terms of use or is flagged for review by other users. However, there is a delay between the time a video was removed from the site and the time it will stop appearing in search results. This can cause data previously accessed to be incorrect—a video may appear as a vertex in the network but will not actually remain a part of it in the current data on YouTube.

13.6.3  Ethical considerations YouTube users can choose to make certain segments of the information on their channel private. They can decide that their subscribers lists are private, approve or delete comments, and send private messages to other users. When importing data from YouTube, it is important to consider that data may be missing some components

13.7  Preparing YouTube network data Using the YouTube video network importer will result in collecting multiple edges for the same vertices (i.e., videos). As a result, it is possible for videos to be connected to each other more than once. When the data

III. Social media network analysis case studies

194

13.  YouTube: Exploring video networks

is imported, you will see the Relationship column on the Edges Worksheet will say Shared commenter, since that is the type of edge. This will create duplicate edges—one for each comment. Even multiple comments from the same user will create duplicate edges (with unique content, but the same videos in the Vertex1 and Vertex2 columns). This is so that the content of each comment can be viewed (see Video1 Comment and Video2 Comment columns). In such cases, you can Count and Merge Duplicate Edges feature (available in the Prepare Data drop-down in the NodeXL Ribbon) present a clearer picture of the relationships, allowing you to visually and statistically represent the data accurately. Clicking on the first two options will compare each edge and find the duplicates and then merge them into one connection. The connection will be weighted according to the number of duplicate relationships that were merged. A column containing the edge weight is added to the NodeXL Edges worksheet.

13.8  Analyzing YouTube networks 13.8.1  User networks YouTube users (or “YouTubers”) are content providers, drawing other users, viewers, and participants to the videos they produce. However, YouTube users do not limit their activity on the site to video content production. The massive amounts of social interaction that occurs on YouTube strengthen the idea that the site is host to a complex and lively community of users and participants and is more than just a platform through which incidental viewers can watch videos [3–4]. YouTube user networks are important for several reasons. (1) Users are the essence of the YouTube community; the way they congregate and interact may reveal the flow of information in the YouTube network and highlight the importance of certain users to the community or to a specific sub-community of users. (2) Differences between users’ networks can demonstrate different participation patterns and aid in improving the interface to accommodate various audiences. (3) These networks can also be used by companies and organizations wishing to use the YouTube platform for advertising, lobbying, or disseminating information. Users’ egocentric networks tell an interesting story about the type of ties created among users who subscribe to both the central user being analyzed, and other users that subscribe to each other. The first network you will analyze in this chapter belongs to the popular YouTube video game content creators “Game Grumps”  (see file link provided above). The “Game Grumps” are a small team of creators whose primary use for YouTube is to

­ pload and share “Let’s Play” videos of themselves u playing video games. To see the network, use the Import From User’s YouTube Network option in the NodeXL ribbon as shown in Figure 13.3. Selecting 1.5 levels will retrieve both users who have ties to the Game Grumps and the ties among themselves. This will allow you not only to analyze users’ relations with the person who is the middle of the egocentric network but also to find any hubs of relationships that exist among the other users. In the Edges worksheet tab you can see each pair of vertices and a new column titled Relationship, which is added to the standard NodeXL Edges worksheet and filled according to the type of YouTube relationship that connects these vertices. A quick look at the relationship column (Figure 13.4) will reveal that there are only Subscribed To relationships between the vertices. Note that in the Vertex 1 and 2 columns you will find User IDs. The information about each user is located in the Vertices spreadsheet, as a user—a vertex—is the unit of analysis. Select the Vertices spreadsheet to learn more about the users in the network. While the Vertex uses the unique YouTube ID of the user, you can find user information in the Title and Description columns. See Figure 13.5. You will return to these spreadsheets as we start calculating social networks metrics and visualize the graphs. For now, you will create a visual display of the egocentric network that surrounds the Game Grumps. It is useful to calculate network graph metrics that will describe the shape of the graph and describe the location of each person or video within it. You can then move on to explore the characteristics of the graphic display of the network. Click on Show Graph and look at the initial visual display (Figure 13.6). Each vertex represents a user in the network that subscribed to Game Grumps. Each edge captures a subscription-relationship between two users. Because this is a directed network, there are arrows indicating which user follows which user. In order to make better sense of the network, you can highlight key vertices, based on their Vertex-level metrics.

FIGURE  13.4  User’s network Edges spreadsheet showing User IDs and a Subscribed To relationship type.

III. Social media network analysis case studies



13.8  Analyzing YouTube networks

195

FIGURE 13.5  User's network the vertices spreadsheet showing the User ID, Title (i.e., username), and Description columns.

FIGURE 13.6  User's network—initial graph.

• Calculate vertex in-degree, out-degree PageRank, betweenness centrality, vertex reciprocated vertex pair ratio, and edge reciprocation using Graph Metrics. • Use Autofill Columns to set Vertex Color to OutDegree, Vertex Size to Degree Centrality and Vertex Label to Title. See Figure 13.7. • Set Vertex Shape to Label, using the Graph Options Pane. It appears on the top-right side of your Graph pane (Document Action). See Figure 13.7. If you wish, you can change other characteristics of Vertices and Edges, such as the Curvature of the edges. • Click Refresh Graph (Figure 13.9). Let us examine the graph (Figure 13.8). Each Vertex is represented by its user, as we designated earlier. Examine the Vertices spreadsheet and find the top out-degree and

betweenness centrality vertices (users). You can use the sort options to organize these columns from largest to smallest. Now, examine the graph. At the center you will find the Seed for this ego-network, the GameGrumps YouTube account. You designated vertex color to be associated with out-degree, so it has the darkest color as it subscribed to the largest number of users in the network (not a surprise, as you collected data about this user’s subscriptions). You set vertex size to correspond with betweenness centrality, and as expected GameGrumps has the highest betweenness centrality as it is connected to all users and they are all connected to others either directly or through this account. Looking at the rest of the users, you can see that users at the core of the network are more connected, their higher number of subscriptions (out-degree) gives them a darker color and their

III. Social media network analysis case studies

196

13.  YouTube: Exploring video networks

FIGURE 13.8  Graph options.

FIGURE 13.7  AutoFill columns.

BroDuo Give Heart Records - Nathan Sharp SoloTravelBlog Em-One

The Game Theorists

Hot Pepper Gaming

Funhaus Commander Holly

Gamer Next Door KittyKatGaming

Rooster Teeth Mortem3r

Markiplier Raleigh Ritchie

SuperMega

Egoraptor

NInja Sex Party

SleepyCabin

RubberNinja

GameGrumps

jacksepticeye

tithinian RubberRoss

DidYouKnowGaming? AbsolutelyAlex

GrumpOut

Master Sword

LazyPillow (Yotam Perel)

oddboy ProJared

The Game Station DashieGames

Brandon Turner

JonTronShow

Atpunk Cinemassacre Smooth McGroove

Sunder

FIGURE 13.9  User's network—customized visualization. III. Social media network analysis case studies



13.8  Analyzing YouTube networks

higher betweenness centrality values determines their larger sizes. In contrast, at the periphery you will find the smaller and brighter vertices. The graph and the spreadsheets are linked, so if you select a Vertex in the Vertices spreadsheet it will highlight the corresponding vetext in the graph and vice versa. You may consider removing GameGrumps from the network, since including it can obscure connections between others in the graph. This will lose some information (i.e., who subscribes to or is subscribed by GameGrumps), but may be worth it to remove the edges connecting GameGrumps to all other vertices in the network.

13.8.2  Video networks The second layer of networks that can be found on YouTube are the content-related networks that stem from the various kinds of links between videos uploaded on the site. These networks are less about personal affinity or depiction of preexisting relationships and more about shared topicality and thematic association. Understanding video networks can offer you insights about several important happenings on YouTube, from the phenomenon of viral videos to institutionalized information dissemination; how independent and sponsored videos are connected to each other and how users form these connections and react to them. These insights can help others fathom the way YouTube is used for different purposes, and guide their own engagement in this community.

13.8.3  The YouTube “makeup” video network In this section you will focus your attention on an example of videos related to “makeup” and eye shadow (see file link earlier). Videos tagged with the word “makeup” may come from several sources: some are cosmetics companies’ efforts to extend their marketing to reach viewers online, some videos are from makeup professionals who promote themselves by providing tutorials to the masses, others are created by teenage girls who share their first experimentations with cosmetics. “Makeup” is one of the most popular topics on YouTube, with millions of videos related to the topic and an extensive presence in various categories. To start analyzing the YouTube makeup video network, look for separate or overlapping groups or cliques of users create around the shared terms. Some uses of a specific term are distinct from each other (i.e., “relationship” AND “makeup”), whereas other terms blend together (“eye shadow” AND “makeup”). Some videos are clearly personal or amateur, whereas others are professional productions. Start by importing the video network data into the NodeXL workbook using the data shown in Figure 13.2

197

earlier in the chapter. Make sure to check the first box, which will create edges between videos that have comments by the same person. Although NodeXL can import large numbers of videos, because of the popularity of the topic and the huge number of related videos, limit retrieval to 300 videos, 200 for top-level comments and 100 for replies (Figure 13.2). Alternatively, you can download the file linked to earlier from the NodeXL Graph Gallery. Searching for the keyword that you have just typed into the search box, the application will look in all the possible search fields combined. Currently, the YouTube API does not allow for distributed searches that differentiate between various fields (e.g., header, description, tags, and category). Therefore, some of the videos will have the keywords makeup in their header or their description, and some will not. However, all videos (vertices) will have the “makeup” and “eye shadow” keywords. After importing the YouTube makeup video network into NodeXL, the workbook will include the following: an Edges worksheet that includes pairs of vertices along with a column that describes their Relationship (in our case all edges will be connected by Shared commenter), and a Vertices worksheet that will be populated with information about each of the individual vertices (i.e., videos). This includes useful information such as the number of views, comments, likes count, dislikes count, creation date (UTC), title, description, author, tags, and links to the actual videos. Select the Show Graph button in the NodeXL Graph menu to display a visualization of the set of connections among the population of “makeup” videos. The first look at a YouTube video network can be daunting. What can this blob in Figure 13.10 mean? Many shared commenter networks are densely interconnected, since there are people who comment on many of the videos related to a certain topic. Using a range of tools in NodeXL, you can look at deeper layers within this network, filtering obscuring details to reveal interesting things. The first step in this process is to prepare the data, first by using the Count and Merge Duplicate Edges feature in the Import menu (Figure 13.11). In the case of video networks this is especially important as multiple ties can be repeatedly created based on the a­ ctions of only a couple of active commenters. In our ­example, the makeup video network, the Merge Duplicate Edges feature reduced the network from more than 16,000 edges to only 775 unique edges. Note that you are not losing any network data, though you do lose some of the content of the comments in the Other Columns. The number of duplicate edges appears in a new column in the Edges spreadsheet titled Edge Weight. Once the dataset is prepared, it is ready for the creation of network metrics. Compute the metrics relevant

III. Social media network analysis case studies

198

13.  YouTube: Exploring video networks

FIGURE 13.10  YouTube “makeup” AND “eye shadow” video network in its raw form.

FIGURE 13.11  Count and merge duplicate edges.

to undirected networks using the Graph Metrics menu option. Summary metrics about the network are reported on the Overall Metrics worksheet. While an analysis of some of the metrics and the basic visualization can provide some insights, the network is difficult to decipher, and the visualization gives us only a point from which to begin our exploration. Filtering the network, based on Edge Weight, can remove some of the excessive data, especially the peripheral videos that do not belong to the core of the network. Looking at the range of Edge Weights you can see that it is quite wide, ranging from 1 to 8429. This suggests that some pairs of videos have only one shared commenter, while others have many. Remember from Chapter 7 that there are two approaches to filtering in NodeXL: one (Dynamic Filters) operates on the display of vertices

and edges in the visualization pane, and the other operates on the spreadsheet data rows that feed the graph visualization. Unlike Dynamic Filters, when an edge or a vertex is Skipped using Autofill Columns it will not be read into the graph visualization, and clicking on related edges or vertices will not display it. Filtering at the spreadsheet level can be useful for reducing the size of the data sent to computation intensive tasks like the calculation of metrics, clusters, and layouts. Data filtered at this level will never appear in the graph display no matter how the Dynamic Filters are set; you have to be careful about filtering this way and not exclude important parts of the network graph. In this case, using the Autofill Columns option is preferred, as at a later stage you can recalculate the network metrics to include only its core components. Use Autofill Columns to display the wanted resolution of the network. To do that, select Edge Visibility and base it on Edge Weight. In the right-hand Options tab, select an edge weight starting from 2 (Greater than 1) as shown in Figure 13.12. Also map Edge Width to Edge Weight with Options that make the maximum width 2 and use a logarithmic mapping. At this stage you can see multiple isolates that clutter the visualization. An easy way to de-clutter the graph is by applying a different layout for the graph. From the graph pane layout algorithm drop-down menu select Layout Options, and then choose the third option to lay out the smaller connected components at the b ­ ottom

III. Social media network analysis case studies



13.8  Analyzing YouTube networks

199

FIGURE 13.12  Autofill Columns with Edge Visibility Options dialog also open and set to Greater than 1.

of the graph. Then choose the Harel-Koren Fast Multiscale layout algorithm and choose Refresh Graph. This will position the vertices in a much more meaningful way. Once the filtering has presented a workable visualization, the next step is to find clusters (i.e., groups) within the network. You can often recognize clusters in networks visually, but clusters can also be automatically identified as described in Chapter 7. This feature creates a set of groupings of vertices based on their patterns of interconnection. It finds some obvious clusters but can also identify more subtle distinctions that may not be visually obvious. Click on the Groups drop-down menu in the NodeXL ribbon and choose Group by Cluster, selecting the Wakita-Tsurami clustering algorithm from the list and checking the Put all neighborless vertices into one group option. Then, choose Refresh Graph to exhibit the clusters. Look carefully at the clusters that were created: NodeXL assigns each cluster a unique shape or color, and sometimes different clusters may share the same color. If, as in this case, it is hard to differentiate between clusters because of the color similarity, use the Groups worksheet to manually set the group colors or shapes. The resulting graph should be fairly readable. However, you may want to split groups into distinct

boxes by using the Layout Options second choice Lay out each of the graph's groups in its own box (Figure  13.13). Calculate Graph Metrics again, this time including Group metrics. Compare the groups visually on the graph. Next, use the metrics presented on the Groups worksheet to compare them numerically. Notice that some of groups have high density, while others have low density. Navigate to the Vertices worksheet and sort on the Vertex Group column. Then read through the Author, Title, and Description tag to better understand how the videos are related. You can also right-click on any vertex in the graph pane and choose the Play Video in Browser button to open the video in YouTube. This content analysis can help you compare different groups to identify common themes. For example, some groups are based on regions (e.g., India or Pakistan), while others cover introductory tutorials versus more advanced tutorials. Add titles to the groups if desired. It is interesting to discover that no commercial videos or product placement videos were included in this dataset. That may be because of cosmetics companies’ reluctance to use YouTube as an advertising venue or due to the popularity of tutorial videos in comparison to commercial content (a search for “makeup commercial” retrieves only about 4800 results, compared with more

III. Social media network analysis case studies

200

13.  YouTube: Exploring video networks

FIGURE 13.13  YouTube video network displayed by groups with group labels manually added after content analysis of the video descriptions.

than 300,000 results for makeup tutorials and 46,000 for makeup tips), it can also be attributed to the sample size. However, if commercial corporations want to use YouTube as an advertising platform, they should consider how and through which route they can best promote their products. One way is to engage prominent users in product placement (i.e., using specific products in their tutorials or tips). To find the most strategic users to approach, advertisers need to identify the most central users that affect the community. Some users are more influential than others. Social network analysis has several measurements that can be argued to relate to the “influence” of a position in the network. One way you can identify influential users in YouTube networks is by sorting users by their ­betweenness centrality measurement. This value captures the extent to which a person is the only path between two otherwise separate parts of the network. Each person is “between” separate networks, and is a “bridge” between these networks. In YouTube, some videos are linked strategically so that they are the only

connection between separate clusters of videos. At this stage it is important to remember that at the beginning of this analysis you used Autofill Columns to filter all videos with edge weight of less than 2. This not only allowed for a clearer visualization but also affected the accuracy of the network metrics that were previously calculated. Videos that were previously central or prominent may change their position in the network after the filtering. Currently, because you calculated Graph Metrics after filtering, many videos will not have any metrics values. These metrics are thus, focused on the relationships among videos that are more closely connected to one another. Often, when analyzing YouTube video networks, you can identify certain videos that serve as bridges between two clusters. If this is not as apparent, turn to the Vertices worksheet and sort on the Betweenness Centrality column. Select the top row and click or scroll through the list and look at the placement of the high betweenness videos within the clusters. These videos are boundary objects around which the YouTube “eye shadow AND

III. Social media network analysis case studies



201

13.8  Analyzing YouTube networks

makeup” network congregates. Boundary objects are intellectual concepts, artifacts, or objects that connect different communities of interest [2, 5], though each community may interpret or use them differently. Boundary objects are also used as translational mechanisms—they provide a channel for transferring information, ideas, and understandings between different communities, where each community recognizes the boundary object’s common structure yet applies to it its own interpretation. A video is not a boundary object per se—it becomes one when different communities give it meaning and use it as such. It is elastic enough to accommodate the different meanings attributed to it by the different clusters. It is interesting to examine overall placement of the videos in the entire YouTube network. This can be done using the statistical data provided by YouTube (e.g., views, likes count, dislikes count, number of comments), which is also imported into NodeXL. These statistics can help you compare videos central to the entire YouTube social network. What kind of story could you tell if a video has a high betweenness score, but a low number of views compared to less central videos within the

­ etwork? In this dataset, the video that is most central to n the network for linking videos together at a centrality of 1580.169 has less than 95,000 views, while the video with the most views—more than 18 million—has a centrality score of less than 25. Despite having millions of more views than our central-most video, this video is out on the periphery of our network. To better view this metadata, use Autofill Columns to set the vertex Shape to Views and in Vertex Shape Options select Label for every video with more than 3,600,000 views. Set Vertex Size to Degree, Vertex Layout Order to Views, and Vertex Visibility to Degree. Examine the Vertices spreadsheet and find that for the 19 most-viewed videos on our network the Shape column indicates Label, using Vertex Shape Options to include only vertices with Degree Centrality of 2 or more in order to hide the less connected vertices. The result should look something like Figure 13.14. Examining the graph that you just created you will notice that videos and users that have widespread popularity among a general population of viewers (i.e., those showing up as a label instead of a disk) may lack

Denitslava Makeup

Jaclyn Hill

An Knook Everything TV - DIY, Makeup, Hairstyles, Nail Art Maya Mia

ModaMob

Emma Pickles

Tina Yong Thalita Ferraz Christen Dominique

Tina Yong Wayne Goss Laura Lee

Denitslava Makeup

AlexandrasGirlyTalk

Created with NodeXL Pro (http://nodexl.codeplex.com) from the Social Media Research Foundation (http://www.smrfoundation.org)

FIGURE 13.14  Video network with top viewed videos.

III. Social media network analysis case studies

202

13.  YouTube: Exploring video networks

FIGURE 13.15  Filtered YouTube eye makeup tutorial network.

influence within local communities of interest (i.e., they may not have a high degree). You will also notice that of the 19 most-viewed videos only 15 appear in the graph. Look at the Visibility column in the Vertices spreadsheet and note that for the missing 5 videos the Visibility is set to Skip. A quick look at their degree values will reveal that they have very low values (under 2). Since you set the Vertex Shape Options to include only vertices with degree centrality of 2 or more, these high viewed videos are not in the graph. This is another illustration that videos may be very popular on YouTube and still peripheral in your video network. To display complex combinations of network attributes, you can map different metrics to different visual properties. A node may hold different attributes like its centrality in its local network and popularity in the overall YouTube population, measured in terms of the number of views or comments. This is demonstrated in Figure  13.15 where, using a fresh copy of the dataset, edge-bearing vertices were filtered out if they had less than 700 comments, labeled by author, with each node shaped to the video’s thumbnail image.

These additional attributes can help distinguish between different kinds of popularity and activity in YouTube, showing that some videos are popular but do not generate discussion and vice versa. These additional attributes can also help illustrate and reveal interesting insights about our networks.

13.9  Practitioner’s summary Analyzing YouTube social networks can offer many insights into the ways videos become popular, sometimes even becoming “viral,” and the way information is disseminated through videos. YouTube’s popularity makes it a channel that enables professionals, from marketing experts to political advisers, to gauge popular themes and public trends. Analyzing video networks will make it easier to decide on the types of ­interventions, the creative routes that will maximize their outcome, and, even more important, what approaches not to take in order to avoid negative backlash.

III. Social media network analysis case studies



Suggested reading

For artists and content producers who are not backed by large-scale media corporations, simple social network analysis—focusing on egocentric and content networks—can give a glimpse of the many facets that affect success and popularity within the YouTube universe. As YouTube becomes an important tool for i­ nformation dissemination in other noncommercial domains, such as education and public health (e.g., the Center for Disease Control’s streaming channel2), social network analysis of a different kind—one that explores which audiences can be reached through YouTube videos, for example—can provide a useful tool to coordinate funding and production efforts in an efficient manner. These observations, coupled with YouTube’s immense popularity, provide deep insights into one of the major media outlets in existence today. The size of the YouTube network can be daunting, but by focusing on appropriate samples of data, filtering with relevant metrics, and using visualizations extensively, you can grasp at least part of what shapes our contemporary culture.

13.10  Researcher’s agenda Despite YouTube’s immense popularity, the research on YouTube’s underlying social networks is in its early stages. Although practitioners, from marketing experts to educators, have attempted to explore these networks to gain an understanding of the best ways to utilize YouTube for information dissemination, researchers have, for the most part, preferred to study more “obvious” social networks such as Facebook and Twitter. Structural studies of YouTube have focused on its overall macrostructure [6–7] or on a category level [3, 8]. The rich composite data found in YouTube offer a compelling reason to map the networks of connection it contains. The combination of user-generated content and social ties can illuminate many phenomena that shape not just our popular culture [9] but also the way institutional information is disseminated or the ways in which public opinion is broadcasted [10]. Increasingly, it is used to understand propaganda campaigns as well [11]. Using social network analysis, researchers can identify important YouTubers or pivotal videos, as well as the types, structures, and development of the networks that are created around them. Researchers can also explore how the structure of ties and networks on YouTube affect content creation. The interplay between content and structure is one of the more important attributes of YouTube and is worthy of deeper exploration. As you have seen, different users or videos have different networks built around them. Understanding the nature and evolution of these networks can lead to im2

www.youtube.com/user/CDCStreamingHealth.

203

proved use of YouTube by users and enterprises or by designers of video-sharing interaction. Researchers can extend our knowledge of the social processes that underlie the YouTube interaction and the ways the social networks that exist on the site contribute to the popularity of or disregard certain views, opinions, or video content.

References [1] B.  Kane, One Million Followers: How I Built a Massive Social Following in 30 Days, BenBella Books, Dallas, TX, 2018. [2] E.  Wenger, Communities of Practice: Learning, Meaning, and Identity, Cambridge University Press, Cambridge, MA, 1998. [3] D. Rotman, et al., The community is where the rapport is: on sense and structure in the youtube community, in: presented at the Proceedings of the Fourth International Conference on Communities and Technologies, University Park, PA, June 4–7, 2009. [4] D. Rotman, J. Preece, The “WeTube” in YouTube: creating an online community through video sharing, Int. J. Web-based Commun. 6 (2010). [5] S.L. Star, J.R. Griesemer, Institutional ecology, ‘translations’ and boundary objects: amateurs and professionals in Berkeley’s museum of vertebrate zoology, Social Stud. of Sci. 19 (1989) 387–420. [6] M.  Cha, et  al., I tube, you tube, everybody tubes: analyzing the world’s largest user generated content video system, in: Presented at the Proceedings of the 7th ACM SIGCOMM Conference on Internet Measurement, San Diego, CA, October 24–26, 2007. [7] G.  Geisler, S.  Burns, Tagging video: conventions and strategies of the YouTube community, in: Presented at the Proceedings of the 7th ACM/IEEE-CS Joint Conference on Digital Libraries, Vancouver, BC, Canada, June 18–23, 2007. [8] J.C.  Paolillo, Structure and network in the YouTube core, in: Presented at the Proceedings of the 41st Annual Hawaii International Conference on System Sciences, January 07–10, 2008. [9] J.  Burgess, J.  Green, YouTube: Online Video and Participatory Culture, Polity Press, Malden, MA, 2009. [10] V. Gueorguieva, Voters, MySpace, and YouTube: the impact of alternative communication channels on the 2006 election cycle and beyond, Social Sci. Comput. Rev. 26 (2008) 288–300. [11] J.  Klausen, E.T.  Barbieri, A.  Reichlin-Melnick, A.Y.  Zelin, The YouTube Jihadists: a social network analysis of Al-Muhajiroun’s propaganda campaign, Perspect. Terrorism 6 (1) (2012).

Suggested reading M.D. Shoham, A.B. Arora, A. Al-Busaidi, Writing on the wall: an online “community” of YouTube patrons as communication network or cyber-graffiti? in: System Sciences (HICSS), 2013 46th Hawaii International Conference on IEEE, 2013, pp. 3951–3960. S.J.  Park, Y.S.  Lim, H.W.  Park, Comparing Twitter and YouTube networks in information diffusion: the case of the “Occupy Wall Street” movement, Technol. Forecast. Soc. Chang. 95 (2015) 208–217. S. Hai-Jew, Exploring “User,” “Video,” and (Pseudo) multi-mode networks on YouTube with NodeXL, in: Social Media Data Extraction and Content Analysis, 2016, p. 242. W.W.  Xu, J.Y.  Park, J.Y.  Kim, H.W.  Park, Networked cultural diffusion and creation on YouTube: an analysis of YouTube memes, J. Broadcast. Electron. Media 60 (1) (2016) 104–122. W.W.  Xu, J.Y.  Park, H.W.  Park, The networked cultural diffusion of Korean wave, Online Inf. Rev. 39 (1) (2015) 43–60. K. America, The communicative features of online hate in temporary social networks in Twitter and YouTube, Multilingual Margins 2 (2) (2017) 74. A.  Oksanen, J.  Hawdon, P.  Räsänen, Glamorizing rampage online: School shooting fan communities on YouTube, Technol. Soc. 39 (2014) 55–67.

III. Social media network analysis case studies

C H A P T E R

14 Wiki networks: Connections of culture and collaboration Howard T. Welser*, Nina Cesare†, Derek Hansen‡, Md. Mahbub Or Rahman Bhuyan* *Department of Sociology and Anthropology, Ohio University, Athens, OH, United States †School of Public Health, Boston University, Boston, MA, United States ‡IT & Cybersecurity, Brigham Young University, Provo, UT, United States O U T L I N E 14.1 Introduction

205

14.2 Key features of wiki systems

207

14.3 Wiki networks from edit activity 14.3.1 Wiki networks of general interest

209 211

14.4 Using the NodeXL MediaWiki page network importer to access wikipedia networks

211

14.5 Understanding topics through page-to-page connections 14.5.1 Data collection and processing 14.5.2 Identifying key topics across wikipedia language communities

14.7 Choosing the right sample frame for your wiki research

222

212 213

14.8 Practitioner’s summary

222

14.9 Researcher’s agenda

223

214

References

223

14.1 Introduction A wiki is a website anyone can edit, where every page modification is recorded and archived. The first wiki system, the WikiWikiWeb, was invented by Ward Cunningham in 1995 to allow a group to easily and quickly (“wiki” means “quick” in Hawaiian) edit a set of web pages without having to know HTML or deal with moving files back and forth to a web server. In addition to reducing the technical barriers to creating web pages, wikis make it easy for people to collaborate on writing tasks because the technology of the wiki provides separate spaces for people to create content and to discuss issues related to the content they are creating. Because of

Analyzing Social Media Networks with NodeXL https://doi.org/10.1016/B978-0-12-817756-3.00014-5

14.6 Analyzing the structure of discussion page interaction 217 14.6.1 Mapping networks and identifying disputes within the English international whaling commission talk page 219 14.6.2 Identifying productive members of a talk page community 220

their flexible structure, support for discussion, and ease of use, wikis are an important platform for supporting online communities. For example, Wikia, a company that hosts wiki communities, has approximately 400,000 separate communities, and it is only one of many such platforms. This chapter explores how to use the activity within wikis to create social network representations that can help community designers understand who and what is important to a given community, determine whether the community is healthy, and distinguish between different types of contributors. Although invented in 1995, wikis remained relatively unknown until about 2003, when Wikipedia, the online encyclopedia anyone can edit, started to come to

205

© 2020 Elsevier Inc. All rights reserved.

206

14.  Wiki networks: Connections of culture and collaboration

­ rominence. Wikipedia has become the dominant source p for encyclopedic information online, and it is increasingly understood as a social force that challenges traditional notions of authority, expertise, and knowledge construction [1,2]. Many organizations and educational institutions such as PBWiki, WikiSpaces, and Wikia, now use wikis as knowledge repositories for users. Open source projects and technical question-and-­answer (Q&A) communities use wikis for documentation and support. Although most social media tools support information sharing, wikis are unique in their ability to support collaborative content creation and maintenance. Wiki systems are also excellent sources of data for social network analysis [3,4], which can reveal important lessons about how people work together, what factors encourage long-term substantial contributions, and how leaders emerge. Alongside their richness, they are one of the more technically demanding forms of social media networks to extract and work with. Many wikis give rise to large datasets, which must be either parsed or sampled before any analysis can begin. Wikis offer many types of pages and several ways for editors to interact, providing a more complex social context than systems like email, chat, or Twitter. In this respect, wikis occupy a similar social and practical space as social networking sites like Facebook or Twitter. The majority of wiki communities create information repositories about a specific topic. Whereas some wikis like Gardenology or the Marvel Comics Database take the form of an encyclopedia, other wikis encourage many genres of pages such as lists of annotated links, debates, or how-to documents [5]. The flexibility of wikis makes them highly adaptable to the needs of their users. Company employees use internal wikis to share information and coordinate effort, teachers use the wikis to collaboratively create and share lesson plans, cancer patients share information about clinical trials and doctors, intelligence analysts use Intellipedia to share information across agencies, fans of Star Wars use Wookieepedia to make sense of a complex universe, and Minecraft players use Minecraft Wiki to document the game universe and provide design resources. Despite their wide use and potential as data sources, wiki systems constitute a challenging and complex form of online community with numerous modes of interaction and potentially dozens of ways to represent relations in the form of a network graph. Before beginning an investigation of any wiki system, it’s important to address several questions. Which modes of interaction are most valuable for understanding your online community? Which definitions of an edge are consistent with socially meaningful interactions in the system? What time frames of interaction should be used to demarcate the boundaries of your data collection? Because of these questions and because of the technical challenges

a­ ssociated with collecting many kinds of network data, wiki systems should be considered an advanced social media system for both practitioners and researchers. However, careful planning and data management techniques can make wikis one of the most rewarding areas for network analysis of social media. The focus of this chapter is on introducing several exploratory methods for working with data from wikis in a social network analysis framework. There are many types of wiki networks that might be of interest for social network analysis. This chapter begins with a discussion of these network types and illustrates several questions that could be asked of wiki network data. It also describes attributes of wiki systems and addresses how those attributes may reflect the ways in which people interact in wikis. The core of the chapter introduces the use of NodeXL and its built-in data MediaWiki importer to analyze the hyperlink and communication networks of pages across different languages in Wikipedia. While participation in the Wikipedia community allows users to connect with one another, talk, and share edits across geographic and political boundaries, the presence of language-specific Wikipedia communities means that cultural boundaries do exist within this space. Due to the crowdsourced nature of Wikipedia, these boundaries may generate differences in the content and category structure of pages addressing the same topics. The manner in which pages are framed, discussed and linked may vary according to the sociopolitical or cultural context of the author(s), and these identities may be expressed in part through language. Existing research has documented differences in how local and global historical figures are ranked by importance across 24 Wikipedia language editions [6], as well as documenting how language groups can map onto cultural differences across a wide range of different language wikipedias [7]. Similar patterns may be found in pages that address other cultural institutions, from popular fast food brands to political ideologies. Data for this analysis were gathered using NodeXL’s MediaWiki importer. This tool is designed to download one of two types of primary relationships on Wikipedia—revisions and page hyperlinks—and allow researchers to build networks based on these associations. This chapter focuses primarily on page hyperlinks. On Wikipedia, hyperlinks between pages are created to link topics mentioned within the body of an article. For instance, a Wikipedia page on the social networking site Facebook mentions Facebook founder Mark Zuckerberg, and users are able to click on Mark Zuckerberg’s name to directly access his page. Hyperlinks between pages provide an emergent measure of relevance in the eyes of the community of active editors for a given article. This chapter looks for differences in the page-to-page link structure for environmental and resource allocation

III.  Social media network analysis case studies



14.2  Key features of wiki systems

pages across languages, and anticipates that these differences are attributable to the extent that the culture associated with the language is dependent on the natural resource addressed. Specifically, this chapter will focus on pages addressing whaling across Norwegian, Italian, and English language communities, and will note whether there exist differences in the content and structure of associated topics. Differences in the pages associated with each topic represent differences in each culture’s understanding of the topic. Institutions and concepts central to one language community’s network may not be present in the other.

14.2  Key features of wiki systems Your tour begins with some of the key features of wiki systems that support studying the ways people behave in them. Wikis have a more complex structure than most of the other social media network data sources discussed in this book, such as Twitter or email. In particular, wikis keep a history of all editing activity that can be viewed on a per-page or per-contributor basis. They use namespaces to organize different kinds of contributions to the wiki.

207

Wikis are built on the premise that everything is a page: the content itself, discussion of the content, individual contributors, supporting tools like category pages, and even the policies of the community itself are all contributed by users and evolve over time. Wikis also have user accounts, which track activity by each person, allow self-disclosure, and facilitate direct communication between people. This discussion will be based primarily on Wikipedia and the MediaWiki software that implements it; however, many wiki sites use the MediaWiki software, and the everything-is-a-page structure and namespaces that characterize Wikipedia are common to most wikis. Most people’s view of wikis comes from seeing Wikipedia pages that result from a web search. For example, as of this writing, searching for “whale” on Google returns the Wikipedia page for Whale as the first result (Figure 14.1). The obvious feature of this page is that it presents information about the page’s topic. In Wikipedia, this generally takes the form of an encyclopedia article, although as mentioned in the introduction, these pages can provide other forms of information such as lists of links or instructions. History. The visible topic page for any article is just a hint of the activity that happened in the course of ­creating it.

FIGURE 14.1  An English Wikipedia Entry for Whale. This article page from the English-language Wikipedia displays content and illustrates discussion, edit, and history tabs. These tabs are standard to most wiki systems and they provide access to edit records from which edge relationships and attributes can be measured.

III.  Social media network analysis case studies

208

14.  Wiki networks: Connections of culture and collaboration

FIGURE 14.2  The revision history page for the English “Whale” article. Wiki pages have a related history page that depicts the timing of every edit, indicates the editor or IP address responsible for the edit, provides space for a brief description of the edit, and displays links to the state of the page before and after the edit. History pages are important sources of network and attribute data in wiki systems.

In fact, every time someone edits a Wikipedia page, using the Edit source tab (note that this will show up as View source if you are not logged in, as in Figure 14.1), the software remembers who made the edit, why they said they did so, when it happened, and what changed. Clicking the View history tab within an a­ rticle (Figure 14.2) will show a timeline of every change to a page. These edits can be browsed sequentially and are searchable by content and user. This historical information has allowed researchers to study everything from the overall growth of wiki systems [6] to effective patterns of collaboration in these communities [8,9]; it also supports the development of technical tools that make recommendations based on the topics in which people show interest [10]. You have a number of tools and resources for getting the history of a Wikipedia page or set of pages. Open source wiki software platform MediaWiki offers Application Programming Interface (API) endpoints for multiple wikipedia language communities. Wikipedia histories may also be accessed by external scraping tools. Data are also available via database dumps. Namespaces. Namespaces organize different kinds of contributions. Sometimes, editors disagree on exactly what should be on a given page. Rather than continually editing each other’s versions of a page (a phenomenon known as an “edit war”) or marking up the page itself with comments and disagreements (which often happens, for example, when people collaboratively edit a Word document), the MediaWiki software separates

a­rticle creation from article discussion using namespaces. The articles themselves are in the “main” namespace; for each article, there is an associated page in the “talk” namespace where people can talk about the page, ask questions, request clarification, and resolve disagreements without affecting the main page too drastically. For the main content pages of the wiki, clicking on the Talk tab (Figure  14.1) will open the related talk page. Although the underlying tool is the same, these talk pages typically take the form of threaded discussions. Figure 14.3 illustrates how contributors use talk pages to discuss page-specific edit decisions and to communicate normative standards of appropriate editing. Separating discussion from other editing makes it easier to study the specific ways people collaborate and the strategies they use to resolve disagreements [11–13]. The threaded nature of the discussions provides a natural way to create a social network based on Wikipedia activity, by creating ties between people who reply to each other in a discussion [4,14–16] (Chapter  10). However, capturing the reply structure from talk pages can be challenging. Everything is a page. There are many other namespaces as well, including the File namespace, typically used for uploading images, sound files, or videos that are included in a page. The Category and Template namespaces, where contributors can create technical tools that group pages and create reusable snippets that can support everything from providing a standard set

III.  Social media network analysis case studies



14.3  Wiki networks from edit activity

209

FIGURE 14.3  The article talk page for the English “whale” page. This page is used to coordinate decisions about the best contents for the article page. The edits to this page are made by people who have an interest in the content page and are often made by people who actively edit the article page. This page displays suggested rules and guidelines for talk page engagement. Discussions within this page are threaded and separated by section.

of information to be included for a category to welcoming new users quickly; and the Wikipedia namespace, where people create and debate Wikipedia policies [17]. These namespaces further organize and identify the kinds of contributions people make, allowing community designers to understand and support the roles people take on [15,16,18]. The fact that every type of wiki page is treated just like any other wiki page with a history of changes and edits supports the study of the evolution of the community itself. By analyzing networks of communication recorded via these edits, researchers can assess how individuals take on specific roles based on the kinds of contributions they make [15,18] and how the group makes policy and governance decisions [19]. User accounts. Although many wikis allow anonymous access and editing in an effort to reduce the cost of contribution, most regular contributors to wikis create user accounts. The same wiki feature that records the history of edits and changes allows researchers to examine the evolution of a page or study the edit activity of a specific person. Figure 14.4 illustrates a small section of a contributor’s edit log. The User namespace allows registered users to create a page describing themselves, their interests, their skills, and the parts of the wiki they are most involved in, which is a rich resource for understanding the characteristics of individual community members. The User talk namespace allows for direct

user-to-user communication, another natural resource for creating databases representing social network relationships from Wikipedia activity.

14.3  Wiki networks from edit activity There are many interesting ways to analyze Wikipedia based on the history of page edits and interaction of its users. Using social networking tools to analyze the structure of MediaWiki-facilitated networks, however, requires translating editor interactions and page-to-page connections into relationships with predefined entities and ties. Some of these networks are social and require us to operationalize what a tie between users signifies. In Wikipedia, a strong tie between editors is likely when two Wikipedians have shared edits on the same articles, reply to each other on article talk pages, and make edits to the user talk page of the other. Such strong relationships may signify an important collaborative relationship, or it may also signify an ongoing disagreement or dispute. The norms related to documenting edits, as well as the rules of respect and civility, result in qualitative documentation of the meaning of such interactions. For this reason, analysis of interaction edges assumed from edits should be fact checked against the qualitative record.

III.  Social media network analysis case studies

210

14.  Wiki networks: Connections of culture and collaboration

FIGURE 14.4  The revision history of a User talk page for a Wikipedia user. This page reports a partial history of edits made by users. These contribution pages are an important source of information about editors and about how collaboration is managed in wiki spaces.

The network of hyperlinks embedded in an article may also be considered an ego-centric network, albeit one that represents concepts and topics related to an article. The inclusion or exclusion of particular concepts from any article is subject to rules of collaboration and contribution that are fundamental to Wikipedia, summarized by the “Five Pillars,” which follow: Wikipedia is an encyclopedia; Wikipedia is written from a neutral point of view; Wikipedia is free content that anyone can use, edit, and distribute; Wikipedia’s editors should treat each other with respect and civility; and Wikipedia has no firm rules [20] Because the emergent definition of relevance for a topic or issue stems from the Wikipedians involved in editing that article, the manner in which this topic or issue is embedded within a network of other pages likely reflects Wikipedian’s understanding of its content and significance. Though not explicitly social, similar metrics can be used to unpack the “relationships” within this content network. Hyperlinks between pages may indicate a strong conceptual association. Networks are composed of vertices, or entities, that are connected through edges that represent the relationships between them. Both vertices and relationships can have attributes, such as the strength of a tie between vertices, or the length of time a vertex has been part of the network. As highlighted by Carter Butts [21], the challenge for the analyst is to choose vertices, relationships, and attributes that give insight into questions that matter. This section discusses each of these issues in general in the context of wiki systems and gives a number of possible example networks one might choose to construct.

What is a vertex? In many social network studies, vertices represent individual people or groups, companies, or institutions. In a wiki, each user account is a possible vertex for networks based on relationships between wiki system users. These vertices may be analyzed within the context of talk namespaces. However, for some questions, pages, and even categories of pages might be the appropriate choice for the vertices in a social network. For example, if you want to understand connections between the kinds of topics (or products) a community is interested in, making pages or product categories the star of the show is probably the right choice. The examples in this chapter use pages, as well as usernames as vertices. What counts as an edge? If vertices are defined as users who contribute edits to a wiki, then edges may be defined as one of many activities that display some type of interaction between two users. Although there are many types of potential indicators of relationships between users, let us illustrate three common ways to infer edges between users from edits. First, edits from one user on the user talk page of another may be evidence of a directed communication relationship between the two of them. Second, edits on an article talk page may indicate directed “reply” connections. When editor A makes an edit that refers to a prior edit by editor B, it creates an edge from A to B, which creates part of the reply graph (Chapter 10). Third, edits of a content page may indicate an undirected connection indicating a shared interest. It may also indicate social interaction; in fact, editors who edit a page at about the same time often later go on to interact more

III.  Social media network analysis case studies



14.4  Using the NodeXL MediaWiki page network importer to access Wikipedia networks

directly [22,23]. For example, if editor A and editor B have both edited 4 of the same pages, you can create an undirected edge connecting A and B together with and edge weight of 4. This is a type of “affiliation network” where people are connected to one another based on their shared “affiliations” (i.e., pages they have both edited). Edges may also represent hyperlinks between pages (i.e., content). Articles referenced within another Wikipedia page create a directed edge representing a claim of relevance or connection. This network is similar to the network of hyperlinked pages that makes up the World Wide Web, though it is constrained to just the wiki pages. Another type of undirected link between pages can occur when multiple pages are “affiliated” with the same category. For example, if Page A and Page B are both classified under Category 1, Category 2, and Category 3, you could create an edge between the pages with an edge weight of 3. Finally, pages can be “affiliated” with one another based on having the same shared editors. For example, if Page S and Page T are both edited by 5 different people, an undirected edge with an edge weight of 5 would connect them. What attributes are important? Researchers can leverage a variety of content to assign attributes to edges and vertices. A distinguishing feature of a user vertex may be the particular pages that they edit or abstain from editing. Edit activity may also be used to describe user vertices. The content and complexity of user edits can be measured, and the number of major or minor edits a user makes to a page may be indicative of their role within the wiki [15,23]. For edges between user vertices, the timing of an author’s edits relative to the edits of other users may be important. Edits indicative of communication on talk pages are likely to be in close temporal proximity, or to exhibit a structure known as the third turn, where an edit by one editor occurs between edits of another. Using edges to track the adoption of specific editing tools may also lend insight into which users are influential within their network [4]. Page vertices themselves have attributes that can be coded as variables: for instance, the topics or categories under which they fall, and the presence or absence of key hyperlinks.

14.3.1  Wiki networks of general interest Two wiki namespaces lend themselves naturally to defining relationships between people and pages. Edits to the user talk pages by other users are clear signs of communication; shared edits to content pages are indicative of clearly overlapping substantive interests, or at least, areas of shared attention, and discussions on article talk pages indicate both shared interest and, depending on the reply structure, direct communication. Analyzing co-­ editing behavior (the same person editing two ­different

211

pages) provides information about both user and page networks. If the goal is to explore relationships between wiki articles, analyzing page category membership or explicit links in the text of one page to another is a way to achieve this. It is possible to explore other kinds of entities and relationships through Wikipedia namespaces as well. For instance, one might want to focus on relationships between groups (“WikiProjects”), articles, categories of pages, or even policies. Overall, the interactions between users and pages in Wikipedia offer a variety of pathways for exploring networks between people and ideas.

14.4  Using the NodeXL MediaWiki page network importer to access Wikipedia networks One key feature of NodeXL is its ability to download network data directly from wikis that use the MediaWiki software, including Wikipedia. Based on API tools available through MediaWiki, NodeXL Pro provides options to download relationships between networks and editors centered on a specific page within any MediaWiki domain. To access the MediaWiki importer, select the Import tab and scroll to From MediaWiki Page Network. This will open the dialog box shown in Figure 14.5. Given a seed article (e.g., Social_media) and wiki domain (e.g., en.wikipedia.org), this importer can create several types of networks shown in the Network portion of the importer. As described in Section 14.3.1, there are a variety of networks that may be constructed via links and edits within Wikipedia namespaces, and NodeXL integrates an importer for collecting a range of networks from MediWikis [24]. Some of these networks leverage users as vertices, some leverage pages as vertices, and some display links between both. Edges may represent replies, edits or hyperlinks. Selections for network imports include: • User-User Network: Coauthorship This option downloads a specified number of revisions for a seed article (default is 50), finds all users who contributed these edits, identifies which pages the authors have edited, and generates an edge between users if they have edited the same page. • User-User Network: Discussion This option downloads a specified number of revisions for the talk page associated with the seed article, and generates an edge between users if their comments are sequential. • User-User Network: Article trajectory This option downloads a specified number of revisions for the seed article and generates an edge between users if they generate edits consecutively. • User-Article Network: Hyperlink Coauthorship This tool selects articles to which the seed article links and downloads a specified number of revisions for all

III.  Social media network analysis case studies

212

14.  Wiki networks: Connections of culture and collaboration

FIGURE 14.5  The MediaWiki importer dialog box. NodeXL Pro provides a built-in importer tool that allows the researcher to select network information based on users and pages associated with a specific article within any MediaWiki space.

of these articles (default is 50). For each revision, it establishes an edge between the user who generated the revision and the article in which the revision appears. • User-Article Network: Category Coauthorship Similar to user-article networks for hyperlink coauthorship, this tool downloads articles that are in the same category as the seed article, collects a designated number of revisions, and generates an edge between the user who generated the revision and the article in which the revision appears. • Article-Article Network This selection generates networks of articles and hyperlinks between them. It may capture hyperlinked pages that are 1.5 to two degrees separated from the seed article, and generates edges between them. In addition to gathering networks, the MediaWiki allows researchers to specify desired attributes for vertices well. For instance, the resulting network may include data on the gender of the wiki user or the date they joined the Wikipedia community. These factors may be

important to consider when analyzing dynamics such as network centrality or user influence. Additionally, you may limit the number of recent revisions downloaded, or limit to all revisions that occurred after a specified date.

14.5  Understanding topics through page-to-page connections Our first example of wiki network research highlights the MediaWiki importer tool as it applies to hyperlink connections within wiki pages. This example illustrates how hyperlink ties between wiki pages can reveal differences in the development of those wikis and in differences in Wikipedians’ definitions of what is relevant to the discussion of a particular topic. It focuses on an environmental topic that varies in its current and historical significance—whaling—and explores how content linked to this page varies across Wikipedia language communities.

III.  Social media network analysis case studies



213

14.5  Understanding topics through page-to-page connections

This example uses “whaling” as the seedpage and analyzes this within the English, Norwegian (Nynorsk), and Italian wikipedia communities. Whaling has long been recognized as an environmental and economic issue in English speaking countries. While some indigenous communities within English speaking countries still rely on whaling to harvest natural resources, this practice is widely outlawed for the sake of conservation. Whaling is of strong historic and contemporary relevance for Norwegian speakers. Despite global controversy, commercial operations within Norway continue to hunt whales for food and other products. While fishing and maritime activity is important to the economy of Italy, whales were historically a relatively unimportant resource and whaling is not practiced. Pages within language-specific Wikipedia communities are edited according to a standard set of Wikipedia-wide policies and norms, of which the “relevance” policy shapes what to include on a page and what to exclude. Therefore, differences in which pages are linked from the seedpage for an environmental issue will partly reflect the cultural and socially embedded understandings of that issue and related concepts. In addition to the seed page “whaling,” this example also explores “International Whaling Commission.” The International Whaling Commission (IWC) was established under the International Convention for the Regulation of Whaling (ICRW) in Washington, DC, on 2nd December 1946. The preamble to the Convention states that its objective is to provide for the proper conservation of whale stocks and systematic advancement of the whaling industry. Their extended intention is to ensure that “hunts are as humane as possible for the whale and as safe as possible for the hunters” [25]. The Commission also conducts research, co-­ordinates and funds conservation work on many species of cetacean. In 1982, IWC banned all commercial whaling by a moratorium which was objected by some countries like Norway. However, recognizing nutritional, cultural and aboriginal subsistence values of different parts of the world IWC allows controlled whaling in these areas. The pro-whaling nations like Japan, Norway and Iceland accuse the IWC of basing these decisions upon “political and emotional” factors rather than upon scientific knowledge–even its own Scientific Committee concluded that quotas on some species of whale would be sustainable. Given this controversy, the manner in which the three language communities selected address this organization will likely vary significantly.

14.5.1  Data collection and processing This example uses data from three MediaWiki imports with “whaling” as the Seed Article in English,

TABLE 14.1  Relevant Mediawiki importer selections for collecting the wikipedia hyperlink networks using a single page as a seed Field

Value

Explanation

Seed Article

Hvalfangst

Whaling is

Wiki Domain

no.wikipedia. org

This is the larger of the two Norwegian Wikis (80% of the population uses this spelling system)

Article-Article Checked Network

This defines the network edge as a directed hyperlink, where vertex 1 = page with link; vertex 2 = page that is linked to

Levels to include

1.5

Degree 1.5 networks reveal the structure of relationships among the vertices directly connected to the seed vertex

Download

1

Only one copy of the page is needed

Download 09/28/2018 revisions from

Selecting the same collection date for all three pages helps ensure replication

Norweigian (Hvalfangst) and Italian (caccia_alla_ balena) within their respective Wiki Domains (en. wikipedia.org, no.wikipedia.org, and it. wikipedia.org, respectively). Each download used Article-Article Network as the desired network type with the network degree set to 1.5. This facilitates mapping articles connected to the seed article, as well as connections between these articles. The Download revisions from parameter uses the last date of the last documented revision. These search parameters are displayed in Table 14.1, using the Norwegian page as an example. After collecting these networks, it may be helpful to view only the connections between the articles directly linked to the seed page. Removing the seed page’s incident edges allows you to concentrate on the relevance structure among the degree 1.5 edges without distraction by the connection to the seed edge, which is redundant because it is implied by the import criteria. To accomplish this, navigate to the Edges worksheet and select the seed article in the visualized network. It should then be highlighted in the Edges worksheet, where you can set the Visibility to Skip. This will exclude the seed page from the network visualization and will make sure it is not included in graph metrics and clustering. NodeXL’s Graph Metrics tool is used to calculate summary metrics (i.e., total edges, total vertices, vertex in-degree, vertex out-degree), degree centrality, vertex clustering and edge reciprocation (see Figure  14.6). It is important to note that hyperlink networks obtained differ in their level of development; networks are larger for more highly discussed topics. Therefore, simple network metrics like in-degree and out-degree will not necessarily be sufficient measures of centrality.

III.  Social media network analysis case studies

214

14.  Wiki networks: Connections of culture and collaboration

FIGURE  14.6  NodeXL’s Graph Metrics dialog box. This example uses a variety of metrics aimed to capture the overall degree and degree centrality of vertices using the selections highlighted in this image.

However, both pagerank and eigenvector measures of centrality are scaled in a way that makes them comparable across networks. Within each degree 1.5 network, the pages with high eigenvector centrality (undirected) and high PageRank centrality (directed, in-degree) are most central. Due to different perceptions of whaling across communities, it is likely that concepts and issues described by pages that rank high in regard to eigenvector and PageRank centrality will be different within the Italian, English and Norwegian Wikipedia communities. Since this is a directed network, you can focus on those metrics most relevant to directed networks, though undirected network metrics can still be useful to examine.

14.5.2  Identifying key topics across Wikipedia language communities The first step in this analysis involves generating networks that display all articles connected to the seed article (Figures 14.7–14.9). In this visualization, the Color and Size of vertices highlights the pages that were central in the discussion of whaling within each language

context. Autofill Columns was used to set the Color of the vertices reflects their Eigenvector centrality (undirected centrality among central vertices). The orange vertices are least central, purple are mid ranked and the blue are most central when not considering the directionality of the network (i.e., assuming it is undirected). The Size of the vertices is based on their PageRank centrality (directed indegree centrality among central vertices), with the largest vertices having the highest PageRank. The colors and sizing used are within NodeXL’s default range because there were no extreme outliers in the data. Each network displayed uses the Fruchterman-Reingold layout, though the Harel-Koran Fast Multiscale would work well too by placing less central vertices along the edges and pulling the most interconnected toward the center. These visualization selections help illustrate attribute similarity across the three language networks, with some notable differences. In the Italian language wiki article on whaling is framed as a geographic and historical topic, with some literary relevance. The majority of the more central pages refer to countries of the Atlantic such as Iceland, Greenland, France, and the Faroe Islands. Secondary and tertiary pages refer to particular centuries, years, and locations such as Nantucket, and the novel Moby Dick. While the page for the International Whaling Commission is included, it is not associated with other international bodies, treaties, or environmental groups. However, Japan [Giappone] (where whaling remains a contemporary issue) is connected to the IWC page and to Corea del Sur, though these are of secondary centrality. Notably absent are many scientific details such as particular species of whales. The Norwegian wiki community appears to treat management of whaling in the international scene as a much more relevant issue than does the Italian wiki. The year 2008 is significant because important multilateral agreements were established, and the International whaling commission is central to the discussion as are nations with active contemporary participation in the scientific and international debate related to whaling: Japan, Norway, the United States, South Korea, in addition to the Faroe Islands, which played a major historical role in whaling practices. Secondary and tertiary terms include the whaling method of faeroysk grindadrap (a collaborative hunting method where multiple boats surround a whale and drive it toward shore), as well as many locations in the Faroes where whaling is currently authorized. The conversation related to whaling in the Norwegian wiki is not settled, many sections indicate that citations or references are needed. The English language wikipedia defines whaling management and contestation as relevant, as well as scientific description of whale species, with the IWC, Japan, Norway, Greenland, whale watching, as central, along

III.  Social media network analysis case studies

Bédeilhac-et-Aynat golfo di Biscaglia

secolo Ariège IX secolo

XV secolo Normandia

Paese basco

1800 Massachusetts

Spagna

Medioevo

nativi americani

Francia

XVI secolo

XIX secolo oceano Atlantico

Nantucket

isola

Corea del Sud

Moby Dick

Herman Melville

Canada

oceano Pacifico

1986

Islanda

Fær Øer

Ulsan petrolio

Giappone

Groenlandia

Norvegia

Owen Chase

Amitori-shiki

Grindadráp

inuit

aleuti

Carne di balena Commissione internazionale per la caccia alle balene

olio olio di balena

blubber The Cove - La baia dove muoiono i delfini Smeerenburg Categoria:Pesca Created with NodeXL Pro (http://nodexl.codeplex.com) from the Social Media Research Foundation (http://www.smrfoundation.org)

FIGURE 14.7  Italian Wikipedia’s whaling seed page hyperlink network with seed page removed. This network indicates that whaling is a topic of historical and literary significance. Whaling as an international and conservation issue is central to the discussion, as indicated by vertices such as “petrolio” (petroleum) and “Norvegia” (Norway).

Fangst

harpun

grahval

Svend Foyn dampmaskin Tønsberg

1900-tallet

Arendal

Norsk hvalfangst

Jakt

Sør-Korea hvalfangstekspedisjon hvalbât

2008

Sørishavet

Japan

hvalfangststasjon

vågehval USA

hvaler

hvalkokeri

1584

Hvannasund

Den internasjonale hvalfangstkommisjonen

Norge Sandavágur

Nederland

KlaksvÍk Húsavík revejakt

Greenpeace

hvalolje

Norðskáli

England færøysk

Fuglafjørður

Syðrugøta

Færøyene

Tórshavn

Hvalba

Hvalvík

1800-tallet Miðvágur

Tvøroyri Vágur

Vestmanna Færøyenes regjering

Sandur Fámjin

Bøur

FIGURE  14.8  Norwegian Wikipedia’s Whaling seed page hyperlink network with the seed page removed. This network illustrates that Whaling continues to be a contested issue in Norway, as indicated by the centrality of “Internasjonal hvalfangstkommisjon” (International Whaling Commission) and 2008 (an important date when multilateral whaling agreements were established). Because whaling is active in Norway, some central terms relate to specific whaling techniques.

216

14.  Wiki networks: Connections of culture and collaboration

Solor

Lembata

Grenadines

Taiji dolphin drive hunt

bycatch Russian Far East

Bequia Journal of Heredity Saint Vincent and the Grenadines National Oceanic and Atmospheric erd Conservation Society Administration Hvalba Dolphin long-finned pilot whale aboriginal whaling porp cetacean intelligence

Chukotka Autonomous Okrug

BBC

Bangudae Petroglyphs

beluga whale World Wide Fund for Nature factory ship

North Atlantic

BBC News Food and Agriculture Orga Anti-whaling

gray whale tic white-sided dolphin Japan

The New York Times Harpoon

United Nations

fish

muktuk

Gerard van Bohemen

whale meat

Norway

Australia Tasmania International Whaling Commission folk Island Inuit common minke whale Antarctic minke whale Soviet Union whale Minke whale Arctic blubber Cetacea

narwhal

aleen whale International Convention for the Greenland Regulation of Whaling fin whaleNew South Wales whale oil

minke whale

Salluit

humpback whale

whale watching

indigenous peoples Jared Diamond

Ulsan

Southern Ocean

anti-whaling

United States

Bering Strait

pilot whale

Greenpeace

American Experience hunting Whale and Dolphin Conservation

northern right whale

Flensing

Pamilacan

Western Australia

Faroe Islands

Queensland

Quebec bowhead whale Kingdom of Denmark Collapse: How Societies Choose to Fail or Succeed margarine

FIGURE 14.9  English Wikipedia’s whaling seed page hyperlink network with the seed page removed. This network illustrates that whaling is a topic that carries varied significance, both historical, literary, and contemporary in the context of conservation. The larger network also indicates that the whaling page in the English Wikipedia is more fully developed.

with a long list of whale species such as baleen, minke, bowhead, etc. Secondary and tertiary concepts include material (whale oil, harpoon), historical (Bangudae Petroglyphs), and social/political (anti-whaling, Sea Shepherd Conservation Society, United Nations). Similar to the Norwegian network, the International Whaling Commission is highly central, meaning that the English speaking Wikipedia community focuses strongly on whaling as an environmental and political issue. The larger number of vertices is likely also related to the fact that the English page is longer. While these networks are thematically different, there is an interesting structural similarity across them. Investigating graph metrics reveals that the distribution of Eigenvector and Closeness centrality was bimodal across all cases. NodeXL’s Dynamic Filters tool illustrates this discontinuity and helps obtain a better view of the most central articles within each network by narrowing the visualization to include vertices with Eigenvector centrality values that fall just above the lower of the two modal peaks (see Figure 14.10). For the Norwegian network the minimum Eigenvector value was 0.0326, for the Italian network the minimum Eigenvector value

FIGURE 14.10  Dynamic Filter dialog box. The Dynamic Filtering tool allows you to filter out viewable vertices based on specified metrics. This example filters out vertices with Eigenvector Centrality lower than 0.0326, which is just above the lower of the two peaks.

was 0.0296, and for the English network the minimum Eigenvector value was 0.0186. The resulting networks, with vertices matching the Eigenvalue specifications, are displayed in Figure 14.11.

III.  Social media network analysis case studies



14.6  Analyzing the structure of discussion page interaction

217

FIGURE 14.11  Subgraphs that include only vertices with high Eigenvalue centrality for each language. These graphs illustrate key differences in the extent to which the Norwegian and English versus Italian Wikipedia communities address whaling as an international issue.

The high Eigenvalue subgraphs (Figure  14.11) indicate that the English and Norwegian Wikipedian communities both see the IWC as central to the discussion of whaling, while the Italian Wikipedians do not. While the English language is the most highly developed Wikipedia, the development of the Italian and Norwegian Wikipedias is similar to each other, and so the low centrality of the IWC should indicate a real difference in community held notions of relevance, rather than simply lower development of the wiki. While reducing the network illustrations to include only vertices with high eigenvector centrality helps illustrate which pages are most influential in Wikipedians’ understanding of a topic, the presence of less influential pages may also be important. However, viewing the network structure as a whole makes it difficult to view smaller clusters. One way to view and compare these clusters is to generate a series of subgraph images. NodeXL’s Subgraph Images tool generates these automatically. Figure 14.12 displays generated subgraphs for the top 21 vertices in the language group datasets. Each subgraph is saved to a separate file, and files were sorted by image size as a proxy for subgraph degree. Examining these sub-networks reveals which entities and topics are highly embedded within the page structure of these three language contexts, and the extent to which they are linked to broader conversations. They illustrate that international governing bodies are relatively important in both the English and Norwegian Wikis, but not in the Italian Wiki. Overall, the Norwegian Wiki focuses on whaling in the context of the global community (see: mention of the United States, Japan, South Korea) as it relates to environmentalism (see: mention of Greenpeace), whereas the English wiki addresses whaling as an international issue (see: mention of Greenland, Australia), as it relates to native communities and traditions (see: mention of Inuit, aboriginal). Italian Wikipedia also recognizes

the international importance of whaling (see: mention of Greenland, Spain, Norway) and addresses the ­historical significance of the practice (see: mention of 14th century, medieval) but with less focus on native populations than the United States. These graphs illustrate how NodeXL may be used to import data from Wikipedia and map networks of pages that lend key insight into the meaning and significance of a specific topic across cultural contexts. By examining the overall network structure and focusing on degree centrality of articles linked to “whaling” across the English, Norwegian and Italian Wikipedia communities, differences emerge. For Norway, one of three countries in which whaling is legal, the presence of vertices addressing the International Whaling Commission and specific whaling practices make clear the importance and controversy of global regulations. English speaking countries recognize the importance of global whaling regulations, but this topic is largely of historical and indigenous significance. Italy, which does not share Norway or North America's historical and/or current reliance on whaling as a commercial enterprise, is less involved in conversations addressing whaling regulations. Overall, the structures of these pages and the structure of cultural conversations on this topic parallel one another closely.

14.6  Analyzing the structure of discussion page interaction In addition to analyzing the structure of page-page link networks, you may also use the MediaWiki importer to gather and analyze the user discussion networks underlying the creation of these pages. Given that there are clear differences in the meaning and cultural salience of whaling across language categories, there may be differences in the extent to which contributors across languages discuss article edits and manage disagreements.

III.  Social media network analysis case studies

218

14.  Wiki networks: Connections of culture and collaboration

Img-baleen whale.png

Img-Greenland. png

Img-whale watching.png

Img-Faroe Islands.png

Img-International Whaling Commission.png

Img-whale.png

Img-humpback whale.png

Img-fin whale.png

Img-Inuit.png

Img-Cetacea.png

Img-Southern Ocean.png

Img-whale meat.png

Img-Australia. png

Img-bowhead whale.png

Img-Minke whale.png

Img-common minke whale.png

Img-aboriginal whaling.png

Img-Japan.png

Img-Norway.png

Img-pilot whale.png

Img-blubber.png

Img-Den internasjonale hvalfangstkommi sjonen.png

Img-hvaler.png

Img-Færøyene. png

Img-Norge.png

Img-USA.png

Img-Japan.png

Img-Nederland. png

Img-Sør-Korea. png

Img-Tórshavn. png

Img-2008.png

Img-færøysk.png

Img-Hvalba.png

Img-Klaksvik. png

Img-hvalbåt. png

Img-Sørishavet. png

Img-Vågehval. png

Img-Vágur.png

Img-Norsk hvalfangst.png

Img-Tvøroyri.png Img-England.png Img-Greenpeace. png

Img-Canada.png

Img-Francia.png

Img-Norvegia.png

Img-oceano Pacifico.png

Img-XIX secolo.png

Img-Islanda.png

Img-petrolio.png

Img-Spagna.png

Img-isola.png

Img-Normandia. png

Img-1986.png

Img-Carne di balena.png

Img-nativi americani.png

Img-Grindadráp. png

Img-Fær Øer.png Img-Groenlandia. png

Img-XVI secolo.png

Img-Giappone. png

Img-oceano Atlantico.png

Img-Medioevo.png

Img-Commissione internazionale per la caccia alle balene.png

FIGURE 14.12  Most central (eigenvector) vertices related to whaling across English (top set), Norwegian (middle set) and Italian (bottom set) Wikipedia communities. The sub-network for the International Whaling Commission is highlighted in blue. Viewing subnetworks allows you to view which subtopics are most important to the discussion.

III.  Social media network analysis case studies



219

14.6  Analyzing the structure of discussion page interaction

14.6.1  Mapping networks and identifying disputes within the English International Whaling Commission talk page As described in Section  14.4, NodeXL's MediaWiki importer can collect edit history data from article talk pages. The next example uses discussion ­networks related to the English Wikipedia page for the International Whaling Commission, which were collected using the MediaWiki importer. Set the Seed Page to International_Whaling_Commission and the Wiki Domain to https://en.wikipedia.org/ wiki/Talk. Then select Discussion from the UserUser Network to download all users associated with the sequence of comments on the talk page. By default all wiki pages have the potential for a talk page to be established, but these articles lacked talk pages for both the Italian and Norwegian language groups. Because talk pages are important locations for Wikipedians to solve disputes, their absence may s­ uggest that the contributors to the ICW pages in Norway and Italy lack disputes or substantial disagreement. In the Norwegian case, it may be that all contributors approach the issues from the same political side while the Italian contributors see the situation as a third party without a clear agenda. Thus, this analysis focuses only on the English talk page. The current version of the MediaWiki importer for talk pages uses the order of edits on the history page to infer edges. Figure  14.13 displays a default network

based on the user talk page for the English Wikipedia International Whaling Commission page. The Sugiyama layout provides a good starting place for rendering threaded discussions within this visualization, the color orange indicates the presence of the terms “dispute” or “allegation” in the talk page edit text, highlighting the dyads engaged in conversation related to controversial issues. These were identified using the Word and Word Pair Metrics dialog as shown in Figure 14.14. Specifically, the content in the Edit Comment field on the Edges worksheet was used and List 3 was set to include values with the terms “dispute” or “allegation” (see Figure  14.14 coding of word pairs). This layout (Figure  14.13) indicates that Wikipedia user SammytheSeal is a central player within the talk page, and that this user engages in relatively frequent two-way disputes with user matt77 and some directed disputes with user Jonathanmills. User Andrew GalvenizeDavies, a first-degree connection to SammytheSeal, has reached out to both UberScienceNerd and Istvan with disputes or allegations. It is possible that SammytheSeal and Andrew Galvenize-Davies are experts on the topic, and are sensitive to misinformation. This network map is a good start and provides useful information about active reply structures. However, there are key elements that it neglects. For one, because each page is divided into sections and not all comments are threaded, sequential commenting does not necessarily indicate a reply. Related to this, the structure implies that many users reply to themselves. It is unlikely this

InternetArchiveBot NihlusBOT Cyberbot II Fromthehill Carbonrodney SineBot 76.236.150.223

SammytheSeal

AnomieBOT Yobot Arctic.gnomego999 HagermanBot Nil Einne

Wo0dstock79. 03.35.113 Lifebaka

SmackBot Jonathanmills

Matt77

Vapour Yeu Ninje

82.131.210.162 jam Chris huh Pcb21

Epipelagic

Where be the truthek Istvan

Created with NodeXL Pro (http://nodexl.codeplex.com) from the Social Media Research Foundation (http://www.smrfoundation.org)

Andrew-Galvanize-Davies

UberScienceNerd

FIGURE 14.13  The MediaWiki Importer default network for the English talk page for the International Whaling Commission. The Wikimedia importer infers ties based on edit sequence. Orange ties within this network indicate the presence of allegations and disputes. Bots are shown in blue.

III.  Social media network analysis case studies

220

14.  Wiki networks: Connections of culture and collaboration

FIGURE  14.14  Highlighting communication type within talk page communication graphs. Tie characteristics are available in the Edit Comment column of the network data. For this analysis, ties that contain the terms “dispute” or allegation” are highlighted in orange.

is the case, and possible that they simply contributed to different segments of the talk page in sequence. Finally, this structure ignores the roles that bots play within Wikipedia. Sinebot, for instance, adds signatures to comments on talk pages. These bots do not “discuss” topics in the same way as other users. These networks must therefore be restructured to better illustrate which users directly communicate with one another, and how bots may nudge or direct their conversation (Figure 14.15). Mapping a discussion network that more accurately reflects conversation structure involves cross-referencing the talk page and talk page revision history with the default MediaWiki output. To accommodate this change, two new vertices were added: “FellowWikipedians” for posts that initiate a new topical subsection in the talk page, and “Cleanup” which refers to edits to page attributes that are not direct contributions to the discussion (often changes in links, categories and other dependencies between this page and other parts of the wiki). The resulting network (see Figure 14.15) helps clarify important relationships within the talk page network. It separates page maintenance communication from page discussion (see ties directed toward vertex Cleanup), and allows you to view which users initiate discussions within the talk page (see ties directed toward the vertex FellowWikipedians). Within this

network, it appears that user SammytheSeal still engages in disputes—often with user matt77. However, this user is not directly involved in the construction of the talk page because they do not initiate any discussion topics. Reading the content of SammytheSeal’s contributions shows that this user actively resolves disputes and seeks consensus through reference to relevant citations by engaging in deliberative discussion [19,26]. The revision of the network structure depicted in Figure 14.15 also draws attention to an important social role [15] among Wikipedians, the “wiki gnome.” Arctic gnome, Epipelagic, and others linked to cleanup are engaged in small, inobtrusive tasks that need to be done. Similar to some bots, these wikipedians spend some or almost all of their edits repairing links, updating categories, and similar tasks. While some tasks can be standardized enough to become the focus of bots, others require enough situational judgment that the wiki gnome role remains important.

14.6.2  Identifying productive members of a talk page community Given that the MediaWiki importer also captures text describing interactions within the Edit Comment field,

III.  Social media network analysis case studies



14.6  Analyzing the structure of discussion page interaction

221

FIGURE  14.15  An updated visualization of the English International Whaling Commission talk page, edited so that each tie represents a direct response. Orange ties represent edges coded as disputes or allegations. The “FellowWikipedians” vertex indicates that a post is directed to the entire community (for instance, initiating a subsection).

researchers may also incorporate content analysis into studies of talk page network activity. Studies like this offer great ways to combine qualitative analysis with social network visualization and present viable strategies for professionals and researchers interested in teamwork, small groups, and interpersonal interaction. The example provided in Section  14.6.1 highlighted ties that denote “dispute” or “allegation” using orange. Researchers interested in using the MediaWiki importer may also consider using this tool to analyze other conversation properties and user roles. For example, visualizations can use content from the Edit Comment field to highlight which users are most productive within a user talk page. A productive user may be defined as one who frequently engages editorial tasks or initiates page discussions. Web administrators will want to encourage productive discussion to help avoid edit wars or other dynamics that can squelch widespread participation in the wiki. Identifying which users lead through productivity could help facilitate this. The

following example uses user talk edits for the English Wikipedia article on whaling. The first step in this analysis involves generating a network graph similar to what is displayed in Figure  14.16, in which edges are weighted according to the number of replies and vertices are sized according to each user’s centrality to the conversation. Productivity is measured by counting commonly observed word pairs related to editor leadership actions: “major rewrite,” “new section,” “section article,” and “article concise.” After identifying this content, the NodeXL Autofill Columns feature adds these counts to the graph (vertices with higher counts are more blue). The resulting graph helps researchers to visually identify users (like SammytheSeal, Enuja, and Pcb21) who are both active and productive within the construction of the article. Furthermore, it allows one to note whether there are structural similarities regarding where these users are within the broader page network.

III.  Social media network analysis case studies

222

14.  Wiki networks: Connections of culture and collaboration

FIGURE 14.16  Highlight of the English Whaling talk page, edited so size highlights volume of contributions and color (blue) highlights focus of edits on editorial leadership terms.

14.7  Choosing the right sample frame for your wiki research One challenge associated with conducting wiki research involves bounding your sampling frame. The sampling frame describes the set of edits that you chose to include in your sample and from which your units of analysis will be assembled. You will want to define a disciplined rationale for the inclusion of edits in your study. It is also wise to begin with a carefully selected but limited framework because it is easy to become swamped in too much data. The analysis of discussion within the International Whaling Commission talk page included all edits because the page was not excessively large. If it had been, it may have been wise to select a particular temporal range of edits, or focus only on edits for topics with the most directly interactive conversations. Controversial topics may include multiple archives of talk pages, in which case the proper sample frame is likely to focus on a limited set of topics or section headers within the talk page. For the discussion of the IWC talk page, the validity of collaborative role determinations can be improved by expanding the sample frame for key editors. The initial investigation suggested the SammytheSeal was playing a key collaborative role, while Istvan was advocating for a particular political position (as a member of a well known environmental group). An improved sample frame would investigate the edit histories of both Wikipedians, especially their other contributions to

a­ rticle talk pages to see if their pattern of collaboration indicates a consistent social role across multiple topics. The analysis of page link structures used the idea that only those pages that warranted a direct link from the seed page (i.e., whaling article) were the concepts most directly relevant to the discussion. This sample frame can be improved by investigating related concepts such as seal hunting or overfishing, and making predictions about how these concepts are likely to intersect with the cultural boundaries related to the language groups. In general, it is helpful for sample frames to be drawn narrowly and then expanded incrementally. Internal wikis often grow quite large, it is always best to start your data collection with clear, theoretically based justifications for inclusion of data. Too large a dataset may pose problems in terms of the time and resources needed to manipulate and analyze the network graph. Beginning with a manageable and well reasoned sample frame will help your research be interpretable and replicable.

14.8  Practitioner’s summary This chapter illustrated several structural visualization strategies that can be readily applied to any type of wiki: examining the structure of page links as a means of illustrating understanding of a topic, and identifying which contributors communicate and engage in disputes within a particular page. Although practitioners with access to corporate wikis or other

III.  Social media network analysis case studies



References

223

internal datasets may find it easy to formulate research questions, it will typically be much more difficult to identify the best path for answering them. This chapter addresses key steps in overcoming these challenges. First, familiarity with the wiki system in general will help practitioners identify which types of pages and which projects warrant special attention. Second, practitioners should use knowledge of their particular projects, particular populations of ­editors, or time periods of observation to narrow their attention. Clarifying which questions need to be answered will help practitioners identify which edits to use as evidence of meaningful network relationships and which attributes are relevant and should be measured. Both substantive and theoretical insights are crucial for narrowing the sampling frame in ways that will best answer research objectives. For those engaging in exploratory work, NodeXL is a useful tool because it invites the researcher to be creative. Wiki network studies are a rapidly developing area of research. With just a computer and an Internet connection, your may uncover patterns among wiki collaborators or hyperlinked pages never seen before. The studies in this chapter just begin to scratch the surface of what can be done with wiki-edited network data. The building blocks of analysis and measurement throughout this volume can be combined and applied to wikis, and doing so will likely be fruitful. Part of the impetus behind the development of the NodeXL project is to expand the range of researchers and practitioners who would consider using social network analysis techniques in their research. For many years, network analysis software exacted a high price from people wanting to use these tools in terms of technical skill and willingness to learn proprietary software systems. NodeXL as well as browser-based network visualization tools like TouchGraph are helping expand participation in social network analysis to those without programming skills. The accessibility of these tools opens new pathways for established research questions, and, most importantly, turns the lens of network analysis on new questions and subjects. Because wikis are rich in terms of content and interaction, the growth of new research agendas in the social network analysis of wikis will be especially fruitful.

a particular issue. This chapter illustrates how pages related to whaling vary according to how dependent the cultural context is or has been on the resource. However, researchers may also consider assessing other topics that invite controversy—such as vaccination—and assess the extent to which conspiracy influence and/or partisanship is influential in this network. Such work may consider how the structure of pages changes over time, as conversations surrounding a topic evolve and new ideas and issues are added or removed. Wiki data may also provide an unprecedented window into the process and product of distributed collaboration. The cooperation and coordination problems inherent in any large task are, to varying degrees, overcome in different wiki projects. How is this collaboration achieved [27,28]? What are the key conditions that shape differential levels of collaborative success [8]? Online and off, people frequently adopt distinctive patterns of behavior and interaction known as social roles [15,29]. Wiki systems are no exception; in fact, because of the complexity of the system and diversity of the organizational tasks, wikis are likely to give rise to many distinctive social roles [16,18,23]. Finally, there are other possibilities for network analysis using wikis not addressed by these analyses. For instance, wikis are rich settings in which to study the dynamics of diffusion [4,30,31]. The temporal history of wiki edits is retained with high fidelity, and editors have both durable identities and numerous internal contexts for interaction; researchers can trace diffusion processes better in wikis than they can almost anywhere else. Bridging research on diffusion with studies of cooperation, as well as with those of social roles, will be a fruitful direction for further research. The intention of this chapter is to encourage researchers to make greater use of social media data collected from wiki resources because they offer one of the best combinations of socially relevant, accessible and consistent sources of social media data. While some social media systems are short lived, and others may suddenly restrict access, wiki systems like those supported by the Wikimedia foundation offer a reliable social media source for the study of culture and collaboration.

14.9  Researcher’s agenda

[1] Y. Benkler, The Wealth of Networks, Yale University Press, New Haven, CT, 2006. [2] C.  Shirky, Here Comes Everybody: The Power of Organizing Without Organizations, The Penguin Press, New York, 2008. [3] A.  Capocci, V.  Servidio, F.  Colaiori, L.  Buriol, D.  Donato, S. Leonardi, et al., Preferential attachment in the growth of social networks: the case of wikipedia, Phys. Rev. E 74 (2006) 036116. [4] Y.C. Yuan, D. Cosley, H.T. Welser, L. Xia, G. Gay, The diffusion of a task recommendation system to facilitate contributions to an online community, J. Comput. Mediat. Commun. 15 (1) (2009) 32–59.

The ability of NodeXL to increase accessibility of network analysis opens fascinating new pathways for research. As this chapter illustrates, the ability to link pages between wikis and generate a network of references provides valuable information regarding how wiki contributors—and their cultural context—interpret

References

III.  Social media network analysis case studies

224

14.  Wiki networks: Connections of culture and collaboration

[5] D.L. Hansen, Knowledge Sharing, Maintenance, and Use in Online Support Communities, Unpublished Dissertation, University of Michigan, Ann Arbor, Available online at http://hdl.handle. net/2027.42/57608, 2007. [6] J.  Voss, Measuring Wikipedia, in: Proc. 10th Intl. Conf. Intl. Soc. for Scientometrics and Informetrics, July 24–28, Stockholm, Sweden, 2005. [7] Y.H.  Eom, P.  Aragón, D.  Laniado, A.  Kaltenbrunner, S.  Vigna, D.L.  Shepelyansky, Interactions of cultures and top people of Wikipedia from ranking of 24 language editions, PLoS One 10 (3) (2015) e0114825. [8] A. Kittur, R. Kraut, Harnessing the wisdom of crowds in Wikipedia: quality through coordination, in: Proc. of Computer Supported Cooperative Work, Nov. 8–12, San Diego, CA, 2008, pp. 37–46. [9] U.  Brandes, P.  Kenis, J.  Lerner, D.  Van Raaij, Network analysis of collaboration structure in Wikipedia, in: Proc. of the 18th International Conference on World Wide Web, 2009, pp. 731–740. [10] D. Cosley, D. Frankowski, L. Terveen, J. Riedl, SuggestBot: using intelligent task routing to help people find work in Wikipedia, in: Proc. of the 12th International Conference on Intelligent User Interfaces, Jan. 28–31, Honolulu, Hawaii, 2007. [11] A. Kittur, B. Suh, B.A. Pendleton, E.H. Chi, He says, she says: conflict and coordination in Wikipedia, in: Proc. of 25th Annual ACM Conference on Human Factors in Computing Systems, Apr. 28– May 3, San Jose, CA, 2007, pp. 453–462. [12] J.M. Reagle, Good Faith Collaboration: The Culture of Wikipedia, MIT Press, 2010. [13] F.M.W.  Viegas, K.  Dave, Studying cooperation and conflict between authors with history flow visualizations, in: Proc. of the SIGCHI Conference on Human Factors in Computing Systems, Apr. 24–29, Vienna, Austria, 2004, pp. 575–582. [14] D.  Laniado, R.  Tasso, Y.  Volkovich, A.  Kaltenbrunner, When the wikipedians talk: Network and tree structure of wikipedia discussion pages, in: ICWSM, 2011, pp. 177–184. [15] H.T.  Welser, D.  Cosley, G.  Kossinets, A.  Lin, F.  Dokshin, G.  Gay, et al., Finding social roles in Wikipedia, in: Paper Presented at the American Sociological Association Annual Conference, August 1–4, Boston, MA, 2008. [16] E. Gleave, H.T. Welser, M. Smith, T. Lento, A conceptual and operational definition of social role in online community, in: Proc. of the 42nd Hawaii International Conference on Systems Sciences (HICSS), Jan. 5–8, Waikoloa, HI, Computer Society Press, 2009. [17] I.  Beschastnikh, T.  Kriplean, D.W.  McDonald, Wikipeidan self-governance in action: motivating the policy lens, in: Proc. of International Conference on Weblogs and Social Media, Mar. 30– Apr. 2, Seattle, WA, 2008. [18] J. Thom-Santelli, M.J. Muller, D.R. Millen, Social tagging roles: publishers, evangelists, leaders. in: Proc. of the Twenty-Sixth Annual SIGCHI Conference on Human Factors in Computing Systems, Apr. 05–10, Florence, Italy, 2008. https://doi.org/10.1145/1357054.1357215.

[19] L.W. Black, H.T. Welser, J. DeGroot, D. Cosley, Wikipedia is not a democracy: deliberation and policy making in an online community, in: Presented in the Political Communication Division of the International Communication Association Annual Convention, May 22–26, Montreal, Canada, 2008. [20] Wikipedia:Five pillars, https://en.wikipedia.org/wiki/Wikipedia: Five_pillars. [21] C.T. Butts, Revisiting the foundations of network analysis, Science 325 (2009) 414–416. [22] D.  Crandall, D.  Cosley, D.  Huttenlocher, J.  Kleinberg, S.  Suri, Feedback effects between similarity and social influence in on-line communities, in: Proc. of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Aug. 24–27, Las Vegas, NV, Cambridge University Press, New York, 2008. [23] O. Arazy, H. Lifshitz-Assaf, O. Nov, J. Daxenberger, M. Balestra, C. Cheshire, On the” how” and” why” of emergent role behaviors in Wikipedia, in: CSCW, 2017, pp. 2039–2051. [24] B.C. Keegan, A. Ceni, M. A. Smith. Analyzing multi-dimensional networks within MediaWikis. WikiSym ‘13 Proceedings of the 9th International Symposium on Open Collaboration. ACM New York, NY. https://dl.acm.org/citation.cfm?id=2491056 [25] International Whaling Commission. Working group on whale killing methods and welfare issues. How best do we reference an organizationally produced web page, like the following? https:// iwc.int/working-group-on-whale-killing-methods-and-welfareissues retrieved on 11/19/2018. [26] J. Gastil, L.W. Black, Public deliberation as an organizing principle for political communication research, J. Public Deliberation 4 (2008). Article 3. Available at http://services.bePress.com/jpd/ vol4/iss1/art3. [27] A.  Forte, A.  Bruckman, Scaling consensus: increasing decentralization in Wikipedia governance, in: Proc. of 41st Annual Hawaii International Conference on System Sciences (HICSS), January 7–10, Waikoloa, HI, 2008. [28] S.L. Bryant, A. Forte, A. Bruckman, Becoming Wikipedian: transformation of participation in a collaborative online encyclopedia, in: Proc. of ACM GROUP, ‘05, Nov. 6–9, Sanibel Island, FL, 2005, pp. 11–20. [29] H.T. Welser, E. Gleave, D. Fisher, M. Smith, Visualizing the signatures of social roles in online discussion groups, J. Social Struct. 8 (2) (2007). [30] T.  Hecking, L.  Steinert, S.  Leßmann, V.H.  Masías, H.U.  Hoppe, Identifying accelerators of information diffusion across social media channels, in: European Network Intelligence Conference, Sep. 11, Springer, Cham, 2017, pp. 83–93. [31] M.  Teplitskiy, G.  Lu, E.  Duede, Amplifying the impact of open access: Wikipedia and the diffusion of science, J. Assoc. Inf. Sci. Technol. 68 (9) (2017) 2116–2127.

III.  Social media network analysis case studies

Index Note: Page numbers followed by f indicate figures, t indicate tables, and b indicate boxes.

A ABCD network. See Analyzing Big Complex Data (ABCD) network Affiliation network, 37 Aggregate network metrics, 40, 79, 82 Analyzing Big Complex Data (ABCD) network Graph Metrics dialog, 80, 80f overall graph metrics average geodesic distance, 48 connected components, 31, 81f edges with duplicates, 31 graph density, 48 graph type, 31 key properties, 80, 80f maximum edges in connected component, 48 maximum geodesic distance, 48 maximum vertices in connected component, 48 modularity metric, 48 NodeXL version, 49 reciprocated edge ratio, 31 reciprocated vertex pair ratio, 31 self-loops, 31 single-vertex connected components, 31 total edges, 31 unique edges, 31 vertices, 31 Application Programming Interface (API), 33, 153 Article talk pages, 211 Asynchronous threaded conversation, 19t. See also Threaded conversation discussion forums, 18–20 email lists, 18 email messaging, 18 filter bubbles, 20 Q&A sites, 20 reply graph, 18 reply network, 18 Attribute data, 67 Autofill columns, 70b, 70f, 121–123, 195, 198, 199f

B BBS. See Bulletin board systems (BBS) Betweenness centrality, 41, 83, 168–169, 182–183, 200 Bimodal network, 37 Blogging, 19t blogosphere, 22 description, 22 extension bloggers, 22

livestreams, 23 microblogs, 22–23 multimedia blogs, 23 podcasts, 23 textual content, 22 Boolean search, 162 Bulletin board systems (BBS), 151 Burt's theory, 168, 185–186 Business applications, 44

C Centrality metrics. See Vertex-specific network metrics Citizen-generated YouTube videos, 8 Clauset-Newman-Moore algorithm, 119, 171 Cleaning data, email duplicate addresses removal, 136–137, 136f duplicate edges, count and merge, 137–138, 137b, 137–138f self-referring loops, 136, 138b Closeness centrality, 41, 83 Clustering coefficient, 41 Clusters/communities, 170–173, 171f Co-editing behavior, 211 Collaborative authoring, 19t, 22 Common-interest group, 158 Community conversations, 151–152 Community email collections, 132 Community structure algorithms, 106 Computer information networks, 4 Computer-mediated communications, 5–6 Content networks, 188 Content-related metrics, 175–177, 176f Corporate email list, 153–154 Corporate email network, 96 Co-word analysis, 123, 124f CSCW 2018 Twitter network, 102f. See also Twitter connected component, grouping and visualizing, 103, 104f duplicate edges, count and merge, 102b, 103f filtering, 95 dynamic filters, 103–105, 104f self-loops filtering, 101–102 vertex metrics, 105, 105f visibility column, 105, 106f Group by Motif, 109, 109f group layout and labels, 107–109, 108f group properties and metrics, 107, 108f network clustering algorithms, 106, 106b, 107f subgraph images, 109, 110f

225

D Degree centrality, 40, 83 Design framework, social media control of basic elements, 15–16, 16t directed ties, 17 explicit connections, 16 genre of basic elements, 14–15 implicit connections, 16–17 mutual connection, 17 pace of interaction, 14 retention of content, 17–18 size of producer and consumer population, 13–14, 13t undirected connection, 17 weighted connections, 17 Digital culture, 17 Dunbar number, 43–44 Dynamic filters, 98–99, 99f, 103–105, 104f, 198

E eBay, 14–15 Edge filtering technique, 143 Edge lists, 35, 35t Egocentric networks, 36, 122f, 123, 210 Eigenvector centrality, 41, 83–84 Email, 134b asynchronous, 131 broadcast, 131 collection types, 132, 133t community email collections, 132 components, 133–134 database format, 134 data cleaning duplicate addresses removal, 136–137, 136f duplicate edges, count and merge, 137–138, 137b, 137–138f self-referring loops, 136, 138b deduplication, 134 definition, 130–131 Enron email content networks, 146–147 Federal Energy Regulatory Commission (FERC), 146 entity chunking, 134 entity extraction, 134 entity identification, 134 entity resolution, 134 expertise network email graph, 141, 142f filtering content, 135–136 features combination, 136 folders and labels, 136 sender and receiver(s), 135

226 Index Email (Continued) goals and process, 132–133, 133t history, 130–131 identity resolution, 134 information sharing, 130 Internet Message Access Protocol (IMAP), 134 list, 151 medical patients support, 129 message confidentiality, 130 Multipurpose Internet Mail Extensions (MIME) format, 134 named-entity recognition, 134 network visualizations, 130 NodeXL, import data, 135, 135f organizational email collections, 132 corporate directory system, 142 corporate email network data, 142 multiple information systems, 142 network questions, 133 TechABC (see TechABC email network) organizational networks, 130 personal email collections, 132 data cleaning, 138 data filtering, 138–139 graph analysis, 140 graph metrics computation, 139 network questions, 133 NodeXL, import data, 138 sent and received messages, 139b visualization, 138–139, 140f Post Office Protocol (POP), 134 preparation, 134 push technology, 131 Python library Dedupe, 134 services, 130–131 Simple Mail Transfer Protocol (SMTP), 134 social media tools, 129 technical characteristics, 131 threaded conversation, 131 Encyclopedia of Life project, 7–8 Enron email content networks, 146–147 Federal Energy Regulatory Commission (FERC), 146 Enterprise social networks, 5–6 EU Information Society and Economy and Public Services, 8 Eye makeup tutorial network, 202, 202f

F Facebook, 15, 25 API limits, 180 interaction modes, 179 interlocking networks, 179 network datasets, 179 network visualization, 179 organizational networks, fan pages, 179 cluster-level analysis, 183–184 data collection, 180–181, 181f data collection preparation, 180 Edges spreadsheet, 181, 182f issue-based community, 182 NodeXL importer, 181 vertex-level analysis, 182–183 Vertex spreadsheet, 182, 182f

visualization, 184–185, 184–185f public page, 180 sample datasets, 185–186 seed page hyperlink network, 214, 215–216f Federal Communications Commission (FCC) lobbying coalition network, 109–111, 111f Federal Energy Regulatory Commission (FERC), 146 Filtering data, email content, 135–136 features combination, 136 folders and labels, 136 sender and receiver(s), 135 Flickr, 24 Fruchterman-Reingold algorithm, 111

autofill columns, 70b, 70f edge labels, 67, 71 excel formulas, 69b Label Options dialog, 70, 71f tool tips, 68–70 truncate labels, 68 vertex labels, 67–69, 69f Vertex Shape to Label option setting, 70–71, 72f Language-specific Wikipedia communities, 213 LinkedIn, 25 Livestreams, 23 Location and augmented reality apps and games, 27 Location-based groups, 158

G

M

General Data Protection Regulation (GDPR) legislation, 8 GoodReads, 24 Google, 21–22 Google Hangouts, 14 Group options, 101, 101f, 121–123 Groups worksheet, 183 Group Vertices spreadsheet, 183

Idea generation tools, 26 In-degree centrality, 166–167, 182 Information analysis tools, 4 Information sharing, 130 Instant messaging, 14–15 Interlocking social networks, 43 International Network for Social Network Analysis (INSNA), 43 International Whaling Commission (IWC), 213 Internet Message Access Protocol (IMAP), 134 Internet Movie Database (IMDB), 14 Internet Relay Chat (IRC) network, 20 Internet service providers (ISPs), 151 Iowa Electronic Market, 25–26 iStockphoto, 24

“Makeup” video network Autofill columns, 198, 199f betweenness centrality, 200 count and merge duplicate edges, 197, 198f dynamic filters, 198 eye makeup tutorial network, 202, 202f import from, 197 Isolates group, 199 Layout Options, 198–199 network clusters, 199, 200f network graphs, 202 raw graph, 197, 198f Show Graph button, 197 statistical data, 201 top viewed videos, 201, 201f Wakita-Tsurumi clustering algorithm, 199 Marvel cinematic universe network logarithmic mapping, 40, 88f transformation to unimodal networks, 38b visualizing and interpreting metrics, 39–42, 85f, 87f Marvel Comics Database, 206 Massively multiplayer online games (MMOs), 27 Mathematical sociology, 42 MediaWiki, 14–15 Meetup group, 158 Mobile short messaging service (SMS), 21 Multimedia blogs, 23 Multimodal network, 36 Multiplex networks, 37 Multipurpose Internet Mail Extensions (MIME) format, 134

J

N

H Harel-Koren Fast Multiscale algorithm, 58, 58f, 60f, 68–69, 214 Hashtag, 161–162 HITS algorithm, 45 Homophily, 183 Hubs, 170–171

I

Jacob Moreno's early social network diagram, 38, 39f

K Kaggle, 14 Kickstarter, 25–26 Kiva, 25–26

L Labeling attribute data viewing, 68, 68f

National priorities, social media, 7–8 Network analysis, 96, 96f affiliation network, 37 bimodal network, 37 computer-mediated social relationships, 40 computing resources, 45 data collection, 47 data representations, 34–35, 35f, 35t directed edges, 34 egocentric networks, 36 graphs/sociograms, 31–32 machine-readable network datasets, 47

Index 227

maps of transportation networks, 32 multimodal network, 36 multiplex networks, 37 multi-terabyte databases, 47 network datasets, 45 network metrics aggregate network metrics, 40 centrality measures, 39 clustering, 41–42 community detection algorithms, 41–42 density, 39 distinctive network patterns, 42 grouping, 41–42 patterns of connections, 42 social roles, 42 vertex-specific network metrics, 40–41 node-link diagrams, 48, 49f org-chart, 32 participation metrics, 48 renaissance of network research and data, 42–44 research and practitioner landscape, 37–38 science of networks, 44–45 tools for, 47–48 Twitter network, 33, 33f undirected edges, 34 unimodal networks, 36 unweighted edge, 34 vertices, 34 visualizing and analyzing, 44–45 web of interconnections, 32–33 weighted edge, 34 Network metrics ABCD network (see Analyzing Big Complex Data (ABCD) network) aggregate network metrics, 40, 79, 82 centrality measures, 39, 82–84 clustering, 41–42 community detection algorithms, 41–42 CSCW 2018 conference Twitter network directed network metrics, 89–91, 90–91f Edges worksheet, 89 self-loop, 89 text analysis, 92b Time Series metrics, 92, 92f Top Items metrics, 91–92, 91f density, 39 distinctive network patterns, 42 grouping, 41–42 Marvel cinematic universe network logarithmic mapping, 40, 88f transformation to unimodal networks, 38b visualizing and interpreting metrics, 39–42, 85f, 87f patterns of connections, 42 person-level metrics, 79 social roles, 42 vertex-specific network metrics, 40–41 Niche networks, 25 NodeXL files adjusting Fruchterman-Reingold settings, 60–61, 60f automatic layout, 58f, 59–60, 60f download and installation, 56 features, 55

goal, 55 Graph Options dialog, 61, 63f graph pane updation, 61, 62f layout columns, 61, 62f manual layout, 59 NodeXL Basic, 56 NodeXL Pro, 56 operation edge/vertex highlight, 57f, 58 entering data manually, 56, 57f existing NodeXL file opening, 63 Export menu, 63 graph gallery, 63, 65f graph pane, resizing and moving, 58 importing data, 56–58 menu ribbon, 56, 56f new NodeXL file, 56 opening a NodeXL file created on another computer, 63 saving NodeXL files, 63 showing the graph, 58, 58f spreadsheet and graph pane, 56, 57f trusted location, 64 readable network layout heuristics, 59, 60f undirected and directed graph type, 62–63, 64f NodeXL Graph Gallery, 177 NodeXL social media network diagram, 33, 33f Norwegian wiki community, 214

O Online markets and production, 19t financial transactions, 25–26 review sites, 26 user-generated products, 26 Organizational email collections, 132 corporate directory system, 142 corporate email network data, 142 multiple information systems, 142 network questions, 133 TechABC (see TechABC email network) Organizational Network Analysis (ONA), 5 Organizational networks, Facebook fan pages, 179 cluster-level analysis, 183–184 data collection, 180–181, 181f data collection preparation, 180 Edges spreadsheet, 181, 182f issue-based community, 182 NodeXL importer, 181 vertex-level analysis, 182–183 vertex spreadsheet, 182, 182f visualization, 184–185, 184–185f Org-charts, 5, 32 Out-degree centrality, 166–167

P PageRank algorithm, 21–22, 41, 45 PageRank centrality, 41, 83–84 Personal email collections, 132 data cleaning, 138 data filtering, 138–139 graph analysis, 140 graph metrics computation, 139 network questions, 133 NodeXL, import data, 138

sent and received messages, 139b visualization, 138–139, 140f Person-level metrics, 79 Pinterest, 24 Podcasts, 23 Polinode, 63 Post Office Protocol (POP), 134 Private conversations, 150 Public conversations, 150 Push technology, 131 Python library Dedupe, 134

Q Question and Answer (Q&A) sites, 20, 154–155, 156f, 205–206

R Ravelry bi-modal network graph, 158, 159f common-interest group, 158 location-based groups, 158 meetup group, 158 visual properties, 158 Real-time collaborative neighborhood watch and reporting, 7 Reciprocated edge ratio, 31 Reciprocated vertex pair ratio, 31 Reciprocity measures, 183 Reddit, 14 Relational data, 12 Reply network answer people, 154 cascading style sheets (CSS), 154 CSS-D community, 154–155, 156f, 157 discussion people, 154 eigenvector centrality, 157 graph, 152, 152f Harel-Koren Fast Multiscale layout, 154, 155f online community, 154 Q&A network, 154–155, 156f social network analysis, 154 social role metrics, 157, 157b, 157t

S Scale free networks, 45–46 Selective Exposure theoretical framework, 177–178 Semantic networks text-based dataset, 115 Twitter Gardasil HPV word pair network (see Twitter Search Network) user-generated content, 115 Semi-public conversations, 150 Shared documents, 22 Simple Mail Transfer Protocol (SMTP), 134 Six Degrees of Separation, 42–44 SnapChat, 17–18 Social media applications, 5 definition, 12 Social Media Research Foundation, 55 Social networking services, 19t niche networks, 25 professional, 25 social and dating, 25

228 Index Social sharing, 19t bookmarks, news, and books, 24 description, 23 music, 24 photo, images, and art sharing, 24 video and TV, 23–24 Sociotechnical infrastructure, 11–12 Stack Overflow, 15–16 Strava, 47 Sunbelt, 43 Synchronous conversation, 19t audio and video conferencing, 21 chat, 20 instant messaging, 20 text chat, 21 UNIX talk messaging, 20

T TechABC email network bridge-spanning units, 146 communication patterns, 143–145 edge filtering technique, 143 edge list, 143 Enron Corporation's network, 143, 147f high-traffic connections, 143, 144f interdisciplinary projects, 146 Microsoft Excel's Search function, 145 research and non-research units, 143, 145f self-loops, 143 Threaded conversation, 131 Application Programming Interface (API), 153 bimodal networks, 153 community conversations, 151–152 corporate email list, 153–154 definition, 149–150 duplicate email addresses, 153–154 history, public online, 151 Internet service providers (ISPs), 151 limitations, 154 navigation, 150 permanence, 150 post-and-reply threaded message structure, 149 private conversations, 150 properties, 149–150, 150f public conversations, 150 Ravelry (see Ravelry) reply network answer people, 154 cascading style sheets (CSS), 154 CSS-D community, 154–155, 156f, 157 discussion people, 154 eigenvector centrality, 157 graph, 152, 152f Harel-Koren Fast Multiscale layout, 154, 155f online community, 154 Q&A network, 154–155, 156f social network analysis, 154 social role metrics, 157, 157b, 157t semi-public conversations, 150 single authored, 150 statistics and ratings, 149 threads, 150 thread-to-thread network, 153

topics, 150 Top-Level Reply network, 152, 153f user-to-user network, 153 Thread-to-thread network, 153 Tool tips, 68–70 Top-Level Reply network, 152, 153f Twitter, 11–14, 22–23. See also Twitter Search Network Boolean search, 162 content-related metrics, 175–177, 176f conversations, 162 data collection data import sources, 162, 163f NodeXL Twitter data import feature, 162–163, 164f search string, 162–163 groups, 171f Clauset-Newman-Moore algorithm, 171 clusters/communities, 170–171, 171f Edges worksheet, 172 Groups worksheet, 171–172, 172f hubs, 170–171 identification, 172 Vertices worksheet, 172 hashtag, 161–162 issue-specific designated collections, 161–162 network-level metrics Connected Components, 170 connectivity-related characteristics, 169–170 edges with duplicates, 170 graph density, 170 graph reciprocity, 170 self loops, 170 single-vertex connected components, 170 total edges, 170 unique edges, 170 vertices, 170 NodeXL Graph Gallery, 177 raw data layout Edges worksheet, 163–164, 165f Vertices worksheet, 163–166, 166f retweet, 161 Selective Exposure theoretical framework, 177–178 social interaction patterns, 161–162 vertex-level metrics betweenness centrality, 168–169 degree centrality metrics, 166–168, 167–168f metrics array, 166 user reciprocity, 169, 169f visualization cluster-level layout and visual properties, 174–175, 174–175f clusters, 172–173 Graph Pane, 172–173, 173f user-level visual properties, 173–174, 173–174f vertices, 172–173 Twitter Search Network Gardasil HPV word pair network American English language terms, 116 Count and Salience columns, 117, 119f

Decrease Decimal feature, 117 Edges worksheet, 117, 118f HPV vaccines, 115 search term, 115 Sentiment word lists, 117 Skip these words section, 117 stopwords, 116 unicode languages, 116 Vertices worksheet, 116–117 vlookup formula, 118b word and word pair metrics, 116, 116f Word Pairs worksheet, 117, 117f Words worksheet, 116–117 word networks analysis Clauset-Newman-Moore algorithm, 119 Edges worksheet, 121 Group by Cluster feature, 119 with groups, 119, 120f network clusters, 120 Overall Metrics worksheet, 119 social media messages, 120 sub-groups, 120 undirected network metrics, 119 vertex and edge metrics, 119–120 Vertex Group, 120–121, 121f visualization, 121–123, 122f, 124f

U Unimodal networks, 36 Usenet network, 15–16 User-article network, 211–212 User talk pages, 208–209, 210f, 211 User-user network, 153, 211 U.S. Senate voting analysis co-voting network, 96, 97f dynamic filters, 98–99, 99f filtering edges, 96–98 original data files, 96 Senate115.xlsx source file, 96 vertex attribute Group Option dialog, 101, 101f Group Vertice worksheet, 99, 100f Vertex Color, 101 Vertex ID, 99 Vertex Shape, 101 Vertices worksheet, 96 visibility column options, 97b, 98f

V Vertex data, 67 Vertex-specific network metrics, 82 betweenness centrality, 41, 83 closeness centrality, 41, 83 clustering coefficient, 41, 84 degree centrality, 40, 83 eigenvector, 41, 83–84 PageRank centrality, 41, 84 Video sharing systems, 187–188 Viral videos, 188 Virtual reality worlds, 26–27 Visual properties best practices, 70b edges worksheet, 75, 75–76f graph images saving, 76, 77f graph legend, 76, 77f

Index 229

graph options, 76–77 vertices worksheet, 72, 72f color, 72–73, 73f opacity, 75 shape, 73–74 size, 74, 74f visibility, 75 vlookup, 118b

W Wakita-Tsurumi clustering algorithm, 106, 107f, 199 Wiki document, 13 Wiki networks, culture and collaboration, 15, 22 communities, 205 data sources, 206 direct link, 222 edit activity article talk pages, 211 attributes, 211 co-editing behavior, 211 edge, 210–211 ego-centric network, 210 user talk pages, 211 vertex, 210 encyclopedic information online, 205 exploratory methods, 206 features history, 207–208, 208f namespaces, 207–208, 209f user accounts, 207, 209 “User talk” page, 208–209, 210f Whale page, 207, 207f language-specific Wikipedia communities, 206 NodeXL MediaWiki page network importer, 212f article-article network, 212 article trajectory, 211 feature, 211 user-article network, 211–212 user-user network, 211 page modification, 205 page-to-page connections data collection and processing, 213–214, 213t, 214f Dynamic Filter dialog box, 216, 216f Eigenvector centrality, 217, 217f embedded nodes, 216, 218f graph metrics, 216

Harel-Koran Fast Multiscale, 214 International Whaling Commission (IWC), 213 language-specific Wikipedia communities, 213 map networks, 217 MediaWiki importer, 206, 212 Norwegian wiki community, 214 “relevance” policy, 213 seed page hyperlink network, 214, 215–216f Subgraph Images tool, 217 question-and-answer (Q&A) communities, 205–206 sampling frame, 222 social network analysis, 206 structure, discussion page interaction mapping networks, 219–220, 219–221f talk page community, 220–221, 222f time and resources, 222 Word networks analysis, Twitter Gardasil HPV word pair network Clauset-Newman-Moore algorithm, 119 Edges worksheet, 121 Group by Cluster feature, 119 with groups, 119, 120f network clusters, 120 Overall Metrics worksheet, 119 sub-groups, 120 undirected network metrics, 119 vertex and edge metrics, 119–120 Vertex Group, 120–121, 121f visualization Autofill Columns, 121–123 co-word analysis, 123, 124f ego-network, 122f, 123 filtered view, word pairs, 121, 122f Proquest Thesis and Dissertation dataset, 123 Word Pairs, 93 World Wide Web (WWW), 19t, 21–22

Y YouTube, 179 content networks, 188 data importing NodeXL network file, 191–192 pre-prepared data, 193b privacy preferences, 193 problems, 193 user data, 192–193, 192f

video data, 192, 192f Google tools, 188 “makeup” video network Autofill columns, 198, 199f betweenness centrality, 200 count and merge duplicate edges, 197, 198f dynamic filters, 198 eye makeup tutorial network, 202, 202f import from, 197 Isolates group, 199 Layout Options, 198–199 network clusters, 199, 200f network graphs, 202 raw graph, 197, 198f Show Graph button, 197 statistical data, 201 top viewed videos, 201, 201f Wakita-Tsurumi clustering algorithm, 199 metadata, 188 networks users’ networks, 190, 191f video networks, 189–190 online communities, 188 preparation, 193–194 services, 188 social affinity networks, 188 social network analysis data collection and analysis, 190 user networks, 191 video networks, 190–191 social structures, 187–188 social tools, 188 structure features, 189 network analysis, 189 user channel, 189 videos, 189 user’s network analysis Autofill columns, 195, 196f customized visualization, 195–197, 196f edges spreadsheet, 194, 194f egocentric networks, 194 GameGrumps, 194–197 Graph options, 195, 196f vertices spreadsheet, 194, 195f video content production, 194 video content, advantages, 187 video networks analysis, 197 video sharing systems, 187–188 viral videos, 188