Bayesian Non-Negative Matrix Factorization for Analyzing Co-Location Networks

• Activity Number: 477 - SPEED: Bayesian Methods and Applications in the Life and Social Sciences
• Type: Contributed
• Date/Time: Wednesday, August 1, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Wenna Xi* and Catherine Calder and Christopher Browning
• Companies: The Ohio State University and The Ohio State University and The Ohio State University
• Keywords: Bayesian; Networks Analysis; Sociology; Activity Patterns

Abstract

Co-location networks are a special case of two-mode networks, or bipartite graphs, where ties can only exist between the two modes: individuals and locations. The goal of analyzing co-location networks is to simultaneously study the "dual" association between individuals and the geographic locations where they spend time. In this presentation, we present a Bayesian hierarchical model that allows us to achieve this goal. Our approach can be viewed as a model-based analogue to non-negative matrix factorization (NMF). In particular, we assume that there exists a collection of latent variables, which define "communities" that simultaneously identify groups of individuals who frequent the same locations and groups of locations that are visited by the same people. We illustrate our model via simulation and using data from the Adolescent Health and Development in Context Study.