CASM: A Deep-Learning Approach for Identifying Collective Action Events

• In Event: Deep Learning in Political Science

Abstract

Protest event analysis has been an important method for the study of collective action and social movements. Protest event analysis
typically draws on traditional media reports of collective action, but in some settings, such as authoritarian regimes, government
censorship of traditional media has limited our ability to study protest and mobilization. In this paper, we create Collective Action from
Social Media (CASM)---a system that uses convolutional neural networks and recurrent neural networks to identify collective action events
occurring in the real world with text and image data from social media. We apply our system to China, and demonstrate strong internal
performance as well as external validity compared to existing newspaper-based and hand-curated protest event datasets. We identify 197,734
unique collective action events from 2010 to 2017, creating one of the largest datasets of collective action events in any authoritarian
regime. We discuss the advantages and limitations of using social media for protest event analysis, and we consider how validation can
help make computer science methods more usable for social science research.

Authors

• Jennifer Pan, Stanford University

• Han Zhang, Princeton University