Semi-supervised Latent Block Model with pairwise constraints

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. The Latent Block Model (LBM) is probabilistic model co-clustering based on mixture models that proven useful broad class In this paper, we propose to leverage prior knowledge in the form pairwise semi-supervision row and column space improve algorithms derived from model. We present general framework incorporating must link cannot relationships LBM Hidden Markov Random Fields. instantiate count two inference Variational Classification EM. Extensive experiments simulated real-world attributed networks confirm interest our approach demonstrate effectiveness algorithms.