Deterrence and Censorship

• In Event: Automated Image Analysis in Practice

Abstract

A growing number of studies show how government attempts at online censorship backfire, intensifying rather than suppressing interest in censored topics. The bulk of these studies are based on high profile occurrences of censorship such as internet shutdowns and blocks of popular platforms. In this paper, we use a unique dataset and computer vision methods to examine the effects of user encounters with routine and everyday censorship---for example, finding one link censored while browsing hundreds of links on a social media platform. To capture encounters with routine censorship, we use a custom-built application to passively capture, encrypt, and upload a screenshot image every five seconds from the smartphones of voluntary research participants in China over a week-long period. We then use deep learning techniques to classify mobile activity and identify encounters with censorship on the resulting dataset of 1.5 million images. In contrast to current understandings of the effects of online censorship, we find that routine censorship deters interest in censored topics, even when encounters with censorship generate frustration and anger.

Authors

• Jennifer Pan, Stanford University

• Yingdan Lu, Stanford University