Mapping Twitter Topic Networks:
From Polarized Crowds to Community Clusters
The paper documents the distinct patterns of connection that emerge when people talk to one another using social media services like Twitter. The paper includes six network visualizations that clearly demonstrate the diverse ways people connect to people when using online tools.
This study integrates network and content analyses to examine exposure to cross-ideological political views on Twitter. We mapped the Twitter networks of 10 controversial political topics, discovered clusters – subgroups of highly self-connected users – and coded messages and links in them for political orientation. We found that Twitter users are unlikely to be exposed to cross-ideological content from the clusters of users they followed, as these were usually politically homogeneous. Links pointed at grassroots web pages (e.g.: blogs) more frequently than traditional media websites. Liberal messages, however, were more likely to link to traditional media. Last, we found that more specific topics of controversy had both conservative and liberal clusters, while in broader topics, dominant clusters reflected conservative sentiment.
This is a “pinwheel” diagram using the author’s Facebook personal network (captured July 15, 2009).
Nodes represent the author’s friends and links represent friendships among them. The author is not shown. Each ‘wing’ radiating outwards is a partition using a greedy community detection algorithm (Wakita and Tsurumi, 2007). Wings are manually labelled. Node ordering within each wing is based on degree. Node color and size is also based on degree. Nodes position is based on a polar coordinate system: each node is on an equal angle of n/360º with a radius being a log-scaled measure of betweenness. Higher values are closer to the center indicating a sort of cross-partition ‘gravity’.
This layout has several notable features:
– The angle of each wing is proportionate to its share of the network. Thus 25 percent of nodes go from 0 to 90º.
– Partitions are distinguished by their position rather than a node’s color or shape.
– The tail indicates the periphery of each partition. A wing with many tail nodes indicates many people who are only tied to other group members.
– Edges crossing the center show between-partition connections. Since nodes are sorted by degree it is easy to see if edges originate from the most highly connected nodes or the entire partition.
I spoke on June 4th at the Personal Democracy Forum in New York City about what social media network maps can tell us about various political figures and topics.
Political discussions are obviously a major area of social media use. This talk explores the ways social network analysis and visualization can be applied to mapping discussions of political issues and topics. It features a number of NodeXL generated visualizations of twitter crowds and networks that form around topics like the conference hashtag #PDF2010 (and #PDF10) as well as political and current event relevant terms.
I was also interviewed by Deb Berman from JustMeans.com after the presentation to describe the NodeXL project a bit more:
Here are some sample images of NodeXL topic network maps from the talk:
2010 – June – 3 – NodeXL – Twitter – PDF2010: This map represents the connections among people who tweeted the term “PDF2010”. It illustrates the people in the “center” and the sub-clusters in the map. People who occupy “bridge” locations are visible as well.
2010 – June 1 – NodeXL – Twitter – #tcot: This is a map of the “Top Conservatives on Twitter” tag. It has a large dominant cluster and a tiny sub group of tcot critics.
2010 – June 1 – NodeXL – Twitter – #p2: The #p2 hashtag is used by “Progressive 2.0” discussions. It features a clear dominant cluster of supporters and a smaller cluster of skeptics made up largely of conservatives.
Outside the CUNY Graduate Center auditorium during PDF2010.
Clay Shirky’s talk was great: it wove together stories of collective action for good and trivial purposes that framed a call to increase the costs of political activity on the net rather than reduce as a way to improve the impact of contribution rather than their mere scale.
Howard Rheingold’s discussion with Micha Sifry was insightful, focusing on the ways the Internet can lull us into a lack of mindfullness. A Mindfull approach, Howard encourages, is one where we are not as easily pulled into random tangents and drift aimlessly from link to link and click to click.
The paper describes the roles of “discussion cataylsts” who populate political web boards (newsgroups) and start the threads that get people talking! It turns out that only a very few people in a community get to start many threads successfully. Discussion catalysts have a knack for sparking conversations: setting the agenda for the community at large. Discussion people have high “-in-degree”, they get replied to by lots of people, but low “out-degree”, they tend not to reply that much themselves. The people whoreply to discussion catalysts, in contrast, do reply to one another densely. These are the discussion people, a role that will be the focus of a subsequent paper!
JCMC - Discussion Catalysts
Abstract:
This study addresses 3 research questions in the context of online political discussions: What is the distribution of successful topic starting practices, what characterizes the content of large thread-starting messages, and what is the source of that content? A 6-month analysis of almost 40,000 authors in 20 political Usenet newsgroups identified authors who received a disproportionate number of replies. We labeled these authors ‘‘discussion catalysts.’’ Content analysis revealed that 95 percent of discussion catalysts’ messages contained content imported from elsewhere on the web, about 2/3 from traditional news organizations. We conclude that the flow of information from the content creators to the readers and writers continues to be mediated by a few individuals who act as filters and amplifiers.
Previously, we published “Picturing Usenet” in the JCMC, a paper that features several images of information visualizations of threaded discussions and authors over time. That paper was based on an early work with Fernanda Viegas (now at IBM Research, Cambridge, then as an MIT graduate student in the MediaLab interning with me at Microsoft Research in Redmond, Washington).
[2009 – JCMC- Discussion Catalysts – Himelboim, Gleave and Smith]