Information Propagation Trees for Protest Event Prediction

Abstract

Protest event prediction using information propagation from social media is an important but challenging problem. Despite the plethora of research, the implicit relationship between social media information propagation and real-world protest events is unknown. Given some information propagating on social media, how can we tell if a protest event will occur? What features of information propagation are useful and how do these features contribute to a pending protest event? In this paper, we address these questions by presenting a novel formalized propagation tree model that captures relevant protest information propagating as precursors to protest events. We present a viewpoint of information propagation as trees which captures both temporal and structural aspects of information propagation. We construct and extract structural and temporal features daily from propagation trees. We develop a matching scheme that maps daily feature values to protest events. Finally, we build a robust prediction model that leverages propagation tree features for protest event prediction. Extensive experiments conducted on Twitter datasets across states in Australia show that our model outperforms existing state-of-the-art prediction models with an accuracy of up to 89% and F1-score of 0.84. We also provide insights on the interpretability of our features to real-world protest events.