Title: Modeling Topic-Partitioned Network Structure

Abstract: 

In this talk, I will discuss two projects centered around modeling
topic-partitioned network structure. The first focuses on obtaining
and analyzing local government email corpora. I will describe a field
experiment that we conducted to investigate whether governments'
compliance with public records requests is influenced by the knowledge
that their peers have already complied. I will then talk about
studying local government organizations using the email data that we
obtained via this experiment. I will present a new statistical topic
model intended for discovering and summarizing topic-specific
communication subnetworks. Focusing on the relationship between gender
and professional communication networks, I will explain how this model
can be used to explore persistent topic-specific network
structures. The second project focuses on analyzing country--country
interaction events of the form "country i took action a toward country
j at time t." I will present a new Bayesian Poisson tensor
factorization model, based on the Tucker decomposition, that learns
latent country--community memberships, including the number of latent
communities, as well as directed community--community interaction
networks that are specific to "topics" of related actions types. I
will show that our model infers interpretable communities and network
structures that conform to our knowledge of international relations.