Toward Automatic Censorship Detection in Microblogs

Social media is an area where users often experience censorship through a variety of means such as the restriction of search terms or active and retroactive deletion of messages. In this paper we examine the feasibility of automatically detecting censorship of microblogs. We use a network growing model to simulate discussion over a microblog follow network and compare two censorship strategies to simulate varying levels of message deletion. Using topological features extracted from the resulting graphs, a classifier is trained to detect whether or not a given communication graph has been censored. The results show that censorship detection is feasible under empirically measured levels of message deletion. The proposed framework can enable automated censorship measurement and tracking, which, when combined with aggregated citizen reports of censorship, can allow users to make informed decisions about online communication habits.