Model‐Based Co‐Clustering
Gérard Govaert, Mohamed Nadif · 2013
This chapter talks about the development of the latent block model (LBM) and three algorithms so as to estimate the parameters of this model and leading to co-clustering. The first section talks about developing a general framework for the metric approach for the partitioning clustering situation. Next, the use of adapted probabilistic co-clustering models is described. The LBM is discussed to embed the co-clustering in a probabilistic framework. The chapter demonstrates two types of approximations, namely, a variational expectation-maximization (EM) approach and a classification EM approach to derive the maximum likelihood estimate of a parameter. In the LBM setting, Bayesian inference could be considered as useful in avoiding spurious solutions and thus attenuate the “empty cluster” problem.