Can Bayesian poisson tensor factorization automatically extract interesting events from massive media reports?

Characterization of the dynamical evolution of important events around the globe has become increasingly important. For decades, political scientists have been collecting and analyzing political events of the form “country i took action a toward country j at time t” — known as dyadic events — in order to form and test theories of international relations, respond to crises of all kinds, among others. By constructing daily matrices from a massive political events data, the Global database of events, location (language), and tone (GDELT), which covers almost all the events occurring in the world since 1979 (so far over 400 million in total), we examine whether Bayesian Poisson Tensor Factorization (BPTF) can automatically extract events of interest, such as Paris terror attack 2015, from the sea of events collected by GDELT. For this purpose, we take all the news data in a period of two months from June to August 2016 as a sample, decompose it into 50 components automatically with BPTF. We find BPTF has largely successfully decomposed the mixed events into distinct components, including the conflicts between Palestine and Israel, violent attacks in Germany, Turkey and France, and the South China Sea Arbitration.