An Unsupervised Narrative Detection Framework to Demystify Online Protest Composition

Abstract

In recent years, social media, such as Twitter, has become a central point for organizing and developing online protests worldwide. Although protest and mass mobilization are scarce, they may lead to dramatic developments when they occur. Different narratives shared during a protest serve as strategic tools for building and advancing collective opinions. Hence, it becomes crucial to decipher various narratives shared during an online protest. In this work, we aim to investigate the shared narratives and examine the evolution and communication around the narratives during the protest. To this end, we propose an unsupervised clustering-based framework to understand the various narratives in an online protest. We contribute novel insights about narratives shared during an online protest by analyzing 4 protests under study. Next, we investigate the evolution of identified converging narratives across the protests. We further identify the most influential participants in a protest and study their contribution to spreading various narratives. Our results suggest that clusters with call-to-action tweets and on-ground activity reporting tweets are common narratives across all protests. The analysis of the evolution of narrative suggests that the call-to-action narrative is the most consistent during the protest. The community detection over the retweet network across protests suggests narrative-centric community formation. Our research on narrative analysis during a protest can help provide crucial intelligence and situational awareness to make informed decisions.