id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_yxoxfdikkbbebpsn42ayoshcmu	Ali Mert Ertugrul	Activism via attention: interpretable spatiotemporal learning to forecast protest activities	2019	26	.pdf	application/pdf	13211	1020	58	The theoretically-relevant questions include: in a movement, what social and activity features are associated with the subsequent events? inter-regions are modeled using LSTM (c), with sparse feature learning using Group Lasso (d) This is the first model that differentiates the intraand inter-region contributions in the spatiotemporal event forecasting domain. dynamic features, to a protest indicator at the future time t∗ for a target location d. Dynamic features are to capture social media users' online activities that may be predictive of offline protests. They build event forecasting models for different locations simultaneously by restricting all locations to select a common set of features. all of the three protest events, online activism within a state is predictive of future offline ActAttn enables us to explore the proportion of local (intra-region) and global (interregion) contributions in forecasting protest events, and allows for discovering the "hubs"	./cache/work_yxoxfdikkbbebpsn42ayoshcmu.pdf	./txt/work_yxoxfdikkbbebpsn42ayoshcmu.txt