Photographer: Don Emmert/AFP/GettyImages

Editor’s note: This is the fifth in a series of posts drawn from a new Journal of Peace Research special issue on “Communication, Technology, and Political Conflict.”  The entire special issue has been made available by Sage Publications here.

Policymakers and pundits have argued that social media (such as Twitter, Facebook, YouTube) represents a “liberation technology,” affording new opportunities for people to connect and challenge illiberal regimes or policies. Others have suggested that social media may not be a panacea at all, and in fact allows for more efficient monitoring of dissidents by repressive governments. However, the focus on mass protest and repression obscures the fact that political elites, advocacy groups, and world leaders have increasingly embraced social media.

Scholars have increasingly sought to use the vast amounts of data afforded by social media sources to construct conflict datamap political ideology and polarization, and understand state censorship decisions. In our recent paper in the new special issue of the Journal of Peace Research, we focus on a different, but equally important set of questions—what do the structures of foreign policy discussions on social media look like? Do online networks reflect offline behavior? And, what are the limited inferences that can be made with such data? Our findings suggest that researchers should not treat social media data as a random sample that approximates the general population. Rather, they should use the networked and dynamic nature of the data to answer new questions–e.g. what are the online divisions over foreign policy?

We look at the debate on social media between Israel and Iran over Iran’s nuclear program in 2012-2013. Using key word searches of terms related to the Israel, Iran, and nuclear weapons, we created a dataset of English-language Twitter users engaged in discussions about Israel and Iran. We focused on a subset of users who consistently tweeted about Israel and Iran (so-called ‘super-users’). We then use network analysis methods to look at which individuals cluster together to form communities (share a high number of follow relationships), and which groups of users, or communities are in conversations with each other.

In Figure 1 we show the network graph for the English-language super-users. Each user is represented by a node; the size of the node is proportional to its number of followers. Each color represents a community of users who are tightly interconnected. The distance between communities is also informative, with communities further apart from each other sharing fewer follower relationships. The graph shows five dominant communities: mainstream news media (e.g. @nytimes; turquoise), U.S. liberals/ progressives (e.g. @LOLGOP; dark blue), U.S. conservatives (e.g. @DRUDGE_REPORT; red), Israel supporters (e.g. @IDFSpokesperson; green), and Israel critics (@intifada; purple).

The graph shows that social media relationships of the super users reflect cohesive communities, and that the distance (number of shared followers on social media) between these communities reflects meaningful policy distance. U.S. liberals (dark blue) and Israel critics (purple) are on one end of the graph, with U.S. conservatives (red) and Israel supporters (green) on the end, and mainstream in the middle (turquoise). Furthermore, the graph shows that on the Israel-Iran issue, U.S. conservatives and U.S. liberals are even further apart (as measured by their shared number of followers) than Israel critics and supporters.

Figure 1: Network graph of English-language ‘super-users’

To see if the Hebrew and Farsi social media networks also reflected meaningful policy positions on the Israel-Iran issue, we construct a similar network graph of users and their follower relationships among Hebrew, Farsi, and English Pro-Israel/Israel supporters (for comparison) Twitter users. The results are shown in Figure 2. Not surprisingly, there is a strong division by language, with Farsi Twitter users clustered on one side, and Hebrew users on the other, and very little overlap. Much of the action is within language clusters. In the Farsi cluster we see the main axis of political competition between pro-regime (red), opposition users (green), and a small subset of supporters of the exiled MEK opposition movement (purple). Conversely, the Hebrew cluster is not organized along the main left-right security dimension of Israeli politics, but rather most accounts are personal (pink), focused on entertainment (dark purple), with a few representing dovish/ human rights supporters (light blue).

Many researchers hope to harness the large quantity of social media data to understand international politics. However, they must be cognizant of the differences in network structures across languages. The English-language Twitter network contains many individual, organizational, and professional accounts that reflect the underlying preferences of these actors on the Israel-Iran nuclear issue. Discussion networks in Hebrew and Farsi differ not just from the English network, but also from each other. The Farsi networks reflect the different sides of Iran’s contentious political landscape. Even more interesting, though Iran regularly blocks Twitter, the Farsi Twittersphere still reflects its political divisions. In Hebrew political actors are evident from only one side of the political spectrum (human rights-oriented), and are mainly focused around personal life and Israeli entertainment, not key security issues.

Our findings show that social media actors operate in structured networks that reflect and reinforce communities that form around policy areas. These networks can reveal important information about the politics surrounding important foreign policy issues. But they also show the fallacy of simply using social media data without context.

Figure 2: Farsi and Hebrew Twitter network graphs

Thomas Zeitzoff is an assistant professor at American University.

 John Kelly is the CEO and founder of Graphika.

 Gilad Lotan is the chief data scientist at BetaWorks.