Tweetkeeping: Why Politically-Engaged Iranian Accounts Get Suspended on Twitter

• In Event: Censorship, Content Moderation, and Government Accountability

Abstract

Today, social networking sites (SNSs) such as Twitter and Facebook are entangled in global power struggles. In the wake of the Russian interference in the 2016 and 2020 US presidential elections, anecdotal evidence and company reports suggest that Twitter and Facebook have increasingly targeted user accounts that speak in favor of foreign governments engaged in a geopolitical rivalry with the United States. Countermeasures of this kind raise questions about neutrality of SNSs towards freedom of political speech on their platforms. Nevertheless, an empirical test of this argument as well as a comprehensive understanding of the conditions under which private SNSs curb freedom of speech is lacking. We address this gap in knowledge by examining the conditions under which Twitter suspends user accounts interested in Iranian politics. Based on a sample of 601,940 user accounts that follow Iranian media and politicians (and 65,120,890 tweets sent by these users), this paper uses state-of-the-art machine learning techniques to unpack the extent to which the following factors potentially influence the likelihood of account suspension by Twitter: (a) frequency of political messaging, (b) whether or not an account is automated, (c) whether an account messages in coordination with other accounts, (d) whether it uses hateful language, (e) the ideology of the user, and (f) whether an account messages in support of the Iranian government. We find that those that tweet more often about politics are not necessarily more likely to be suspended. Not surprising, automation and coordination are clear predictors of suspension. Interestingly, we find that conservative (Principlist) accounts and those supportive of the Iranian government are indeed more likely to be suspended. The paper not only provides a first set of illuminating findings on a pressing topic but it also advances a set of methods for studying private social media censorship at a large scale.

Authors

• Andreu Casas, Vrije Universiteit Amsterdam

• Mehdi Zamani, University of Amsterdam