Model-based Poisson co-clustering for Attributed Networks

Abstract

Co-clustering aims at simultaneously partitioning both dimensions of a data matrix. It has demonstrated better performances than one-sided clustering for high-dimensional data. In this paper, we focus on the task of attributed network clustering and propose to consider the edges of these networks as a form of pairwise semi-supervision between data points. We present a probabilistic model for co-clustering based on the Poisson Latent Block Model (LBM) and incorporate probabilistic Must Link relationships in the model using Hidden Markov Random Fields (HMRF). We present two inference algorithms based on Variational and Classification EM. Experiments on the task of node clustering in attributed networks confirm the interest of our approach and demonstrate the effectiveness of our algorithms.