Visualizing the Political Discourse on Twitter


Social media play an important role in shaping political discourse in the U.S. and around the world. However, empirical evidence suggests that politically active web users tend to organize into insular, homogenous communities segregated along partisan lines [1, 2].

In its own right, the formation of online communities is not necessarily a serious problem. However, a deliberative democracy relies on a broadly informed public and a healthy ecosystem of competing ideas. The concern is that if politically active individuals can avoid people and information they would not have chosen in advance, their opinions are likely to become increasingly extreme as a result of being exposed to more homogeneous viewpoints and fewer credible opposing opinions [3].

As part of a series of ongoing studies, we have examined two networks of political communication on Twitter, made up of more than 250,000 tweets from the six weeks prior to the 2010 U.S. congressional midterm elections. Using a combination of network clustering algorithms and manually-annotated data we demonstrate that the retweet network exhibits a highly partisan structure, segregating users into two distinct communities of politically likeminded individuals. In contrast, we find that the mention network does not exhibit this kind of partisan divide. Instead, mentions form a bridge between the two communities, resulting in users being exposed to people and information they would not have been likely to choose in advance.

We hypothesize that these network structures result in part from politically motivated individuals annotating tweets with hashtags that target ideologically opposed users. We argue that this process results in users being exposed to content they are not likely to rebroadcast, but to which they may respond using mentions. In a forthcoming article we provide statistical evidence in support of this hypothesis [4].

Everyone’s an Editor

Partisan Composition of Content Streams
This chart shows the relative number of tweets produced by left- and right-leaning users across a variety of popular information streams. Users from both sides of the political divide are able to contribute content that reflects their own political views.

Hashtags on Twitter, like a radio frequency or television channel, identify content streams associated with different topics and audiences. In contrast to the mass media model, where a single organization can exercise complete editorial control over a stream’s content, hashtags allow anyone to inject their own content into an information stream. Moreover, with hashtag streams, the marginal cost of contributing content to channels with which you might not otherwise engage is almost zero. As a result, a content stream about a well-defined topic, #teaparty for example, can include information that reflects a diversity of views on the subject.

At left we show that for many of the most popular political hashtags, users from both sides of the political spectrum contribute a substantial volume of content. The result is that users who sample information from these streams are likely to be exposed to people, information and opinions with which they might not agree.

Networks of Political Communication

In addition to understanding what people say when they talk about politics on Twitter, one of our primary goals is to understand how people communicate with one another. To this end we collected more than 250,000 tweets containing political hashtags from the six weeks leading up to the 2010 US congressional midterm elections. By recording interactions between users we can create networks of political communication corresponding to the two primary modes of public user-user engagement, retweets and mentions.

Using network clustering algorithms we identified two highly segregated communities of users in the retweet network (below). To understand whether this structure had a meaningful political interpretation we had two of the study’s authors review the tweets produced by 1,000 random users. The authors, working independently, were asked to decide whether the user expressed a ‘left-leaning’, ‘right-leaning’, or ‘undecidable’ political identity in the content of their tweets. To make sure an unbiased party could reproduce our results we compared these annotations with those of a non-author judge, and for a random 200-user subset we report excellent agreement between the authors’ annotations and those of the judge.

These annotations, taken together with the cluster data, reveal a highly partisan structure. In the retweet network, 80% of labeled users in the blue community express a left-leaning political identity, where 93% of labeled users in the red community express a right-leaning identity. In contrast, the mention network does not exhibit this kind of partisan structure, meaning that ideologically opposed users interact much more often using this mode of communication. This difference is particularly important with respect to political communication because it indicates that mentions act as a conduit through which users are exposed to information and opinions that reflect a diversity of political perspectives. Despite these findings, we emphasize that it’s premature to say conclusively whether this inter-ideological communication represents a constructive civil discourse, or whether it’s simply partisan flamebaiting.

Composite Communication Network

The composite of the political mention and retweet networks (7-core shown). Mentions form a communication bridge between the two politically homogeneous retweet communities.

Retweet Network
Among 1,000 manually-annotated users, 93% of users in the red cluster express a right-leaning political identity and 80% of users in the blue cluster express a left-leaning identity. Node colors reflect algorithmically-determined community assignments.

Mention Network

The mention network is dominated by a large politically heterogeneous cluster of users. Compared to retweets, ideologically-opposed users interact with one another much more frequently using mentions. Node colors reflect algorithmically-determined community assignments.

Combining the two networks to form a composite makes it clear that mentions form a bridge between users on the political left and right (shown in the ‘Composite Network’ figure). To explain this we come back to the fact that anyone can contribute content to a hashtag information stream. It’s quite common for users to produce tweets containing hashtags that target multiple politically opposed audiences. For example, consider the following real tweets:

User A: Please follow @Username for an outstanding progressive voice! #p2 #dems #prog #democrats #tcot

User B: Couple Aborts Twin Boys For Being Wrong Gender.. #tcot #christian #tlot #teaparty #p2 #prolife


Each of these users chose to contribute to multiple content streams with primary audiences of likeminded individuals; Progressives 2.0, #democrats, #prog for A, Top Conservatives on Twitter, #christian, #teaparty etc for B. The remarkable thing is that they both also chose to include one hashtag targeting users who would not likely seek out this kind of information on their own. In doing so these users were, with very little effort, able to expose ideologically-opposed consumers of the #p2 and #tcot content streams to their personal political views. Returning to the mass media model, the capacity for content injection creates a situation where literally anyone can decide what’s going to be on TV tonight.

We propose that when a user is exposed to content in this way, she will be unlikely to rebroadcast (retweet) it, but may choose to respond directly to the originator in the form of a mention. Consequently, the network of retweets would exhibit a politically segregated community structure, while the network of mentions would not. In the associated article we present statistical evidence in support of this hypothesis.

Looking Forward

This work is part of an ongoing project at the Center for Complex Networks and Systems Research at Indiana University’s School of Informatics and Computing. Various aspects of this work are slated for publication and release in venues dedicated to the computational, social and political sciences. Concurrent with the International Conference on Weblogs and Social Media we plan to release a network and hashtag dataset based on the information produced during the course of this study. If you have any questions about this or other related works, please contact Michael Conover or any of the other contributors to the project.


[1] Adamic, L., and Glance, N. 2005. The political blogosphere and the 2004 U.S. election: Divided they blog. In Proc. 3rd Intl. Workshop on Link Discovery (LinkKDD), 36–43.
[2] Hargittai, E.; Gallo, J.; and Kane, M. 2007. Cross-ideological discussions among conservative and liberal bloggers. Public Choice 134(1):67–86.
[3] Sunstein, C. R. 2007. 2.0. Princeton University Press.
[4] Conover, M. D.; Ratkiewicz, J.; Francisco, M.; Gonalves, B.; Flammini, A.; and Menczer, F. Political Polarization on Twitter.In Proc. 5th Intl. Conference on Weblogs and Social Media.