Co‐Clustering of Contingency Tables
Gérard Govaert, Mohamed Nadif · 2013
The co-clustering methods have practical importance in a wide variety of applications such as document clustering, where the data are often arranged as two-way contingency tables. In this chapter, two unified frameworks, namely, phi-squared coefficient and latent block model (LBM), are studied. The chapter presents a coherent framework to understand some existing criteria and algorithms for analyzing contingency tables and propose new tables. The first two sections give the necessary background on measures of association used in the chapter. In the third section, the co-clustering approach based on these association measures is presented. The fourth section deals with model-based co-clustering: a block model and an LBM recently developed. Fuzzy and hard co-clustering algorithms are described in detail and some connections between them are established. In the fifth section, the chapter focuses on the comparison and illustrations of the approach presented in the chapter.