Abstract

Transactional data consists of instantaneously occurring observations made on ordered pairs of entities. For example, a set of timestamped emails between coworkers, with one sender and one recipient. Visually, it can be represented as a network, or more specifically, a directed multigraph with edges possessing unique timestamps. In this talk, I explore a Bayesian nonparametric model for discovering latent class-structure in transactional data. By pooling information within clusters of entities, this model can be used to infer the underlying dynamics of the time-series data.