Bayesian Poisson Tensor Factorization for Inferring Multilateral Relations from Sparse.pdfVIP

Bayesian Poisson Tensor Factorization for Inferring Multilateral Relations from Sparse Dyadic Event Counts Aaron Schein College of Information and Computer Sciences University of Massachusetts Amherst aschein@ John Paisley Department of Electrical Engineering Columbia University jpaisley@ David M. Blei Department of Computer Science Department of Statistics Columbia University david.blei@ Hanna Wallach Microsoft Research College of Information and Computer Sciences University of Massachusetts Amherst wallach@ ABSTRACT We present a Bayesian tensor factorization model for infer- ring latent group structures from dynamic pairwise interac- tion patterns. For decades, political scientists have collected and analyzed records 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. We represent these event data as a tensor of counts and develop Bayesian Poisson tensor factorization to infer a low- dimensional, interpretable representation of their salient pat- terns. We demonstrate that our model’s predictive perfor- mance is better than that of standard non-negative tensor factorization methods. We also provide a comparison of our variational updates to their maximum likelihood counter- parts. In doing so, we identify a better way to form point estimates of the latent factors than that typically used in Bayesian Poisson matrix factorization. Finally, we showcase our model as an exploratory analysis tool for political sci- entists. We show that the inferred latent factor matrices capture interpretable multilateral relations that both con- form to and inform our knowledge of international affairs. Categories and Subject Descriptors I.5.1 [Pattern Recognition]: Models; J.4 [Computer Ap- plications]: Social and Behavioral Sciences General Terms Algorithms, Experimentation Keywords Poisson tensor factorization, Bayesian inference, dyadic data, international relations Permission to m