|
Abstract:
|
Many large networks take the form of sequences of interactions between entities, which can be represented as a sparse, structured, dynamically evolving multi-graph. Bayesian edge-exchangeable models allow us to capture appropriate sparsity, and have been incorporated into hierarchical models that are able to capture community-like structure. However, such hierarchical models sacrifice sparsity guarantees, and assume exchangeability of interactions, preventing us from capturing dynamic behavior, such as the tendency of individuals to respond to recent emails. To capture evolving multigraph dynamics, we propose a dynamic Bayesian nonparametric model for interaction networks that tends to reinforce recent behavioral patterns. We describe a time-evolving mixture model that weaves multiple dynamic interaction patterns, creating a nonstationary model that can capture both sparse and dense behavior. We show that the resulting posterior predictive distribution can be used to forecast future interactions, showing impressive predictive performance against a range of state-of-the-art methods. Joint work with Elahe Ghalebi, Hamidreza Mahyar and Graham Taylor.
|
