Activism via attention: interpretable spatiotemporal learning to forecast protest activities

Abstract

The diffusion of new information and communication technologies-social media in particular-has played a key role in social and political activism in recent decades. In this paper, we propose a theory-motivated, spatiotemporal learning approach, ActAttn, that leverages social movement theories and a deep learning framework to examine the relationship between protest events and their social and geographical contexts as reflected in social media discussions. To do so, we introduce a novel predictive framework that incorporates a new design of attentional networks, and which effectively learns the spatiotemporal structure of features. Our approach is not only capable of forecasting the occurrence of future protests, but also provides theory-relevant interpretations-it allows for interpreting what features, from which places, have significant contributions on the protest forecasting model, as well as how they make those contributions. Our experiment results from three movement events indicate that ActAttn achieves superior forecasting performance, with interesting comparisons across the three events that provide insights into these recent movements.

CASTNet: Community-Attentive Spatio-Temporal Networks for Opioid Overdose Forecasting

“…Furthermore, there have been studies that utilized spatial and temporal dependencies for event forecasting and time series prediction. With the success of neural networkbased models, several studies employed neural models to forecast/detect events related to anomalies [4], crime [12] and social movements [5]. Additionally, several studies utilized deep neural models for times series prediction.…”

“…-GeoMAN [17]: is a multi-level attention-based RNN model for spatio-temporal prediction, which shows state-of-the-art performance in the air quality prediction task. -ActAttn [5]: is a hierarchical spatio-temporal predictive framework for social movements. We replace the final classification layer with regression layer to configure it to regression task to use it as another baseline.…”

“…How do locations contribute to communities? CASTNet learns different representation subspaces (communities) of global dynamics unlike the previous work [17,5], and each community consists of a group of different members due to orthogonality penalty. We represent the learned communities and their memberships (i.e., the spatial attention weights α in Eq.…”

“…First, most of them overlook the complex interactions between local (observed from within a region) and global (observed from all regions) activities across time and space. Only a few have paid attention to this problem, yet they model the global activities as a single universal representation [6,5], which is either irrespective of the event location or is reweighed based on a pre-defined fixed proximity matrix [17]. In other words, none of the existing works learns to differentiate the pairwise activity relationships between a particular event location and other locations.…”

Calls to Action on Social Media: Detection, Social Impact, and Censorship Potential

“…Many studies also rely on some combination of spatiotemporal features (e.g. Ertugrul et al (2019); Zhao et al (2015)). The texts of posts could be mined for extracting struc-tured event-related information, or dense meaning representations could be used without identifying specific features, such as doc2vec representations of news articles and social media streams (Ning et al, 2016).…”