Summary
International Symposium on Nonlinear Theory and Its Applications
2016
Session Number:A3L-B
Session:
Number:A3L-B-4
An Alternative to Basic Log-Likelihood for Bayesian Network Clustering
Rei Oshino, Koujin Takeda,
pp.-
Publication Date:2016/11/27
Online ISSN:2188-5079
DOI:10.34385/proc.48.A3L-B-4
PDF download (144.8KB)
Summary:
We study clustering problem of vertices on graph by Bayesian inference. In Bayesian framework of clustering, stochastic block model is a standard for construction of likelihood. Here we start with a variant of stochastic block model by Karrer and Newman for theoretical discussion. It is known that naive log-likelihood from their model does not always give natural clustering. We discuss how to modify it by adding correction term. By numerical experiment, we verify advantage of our method.