Efficient local search in conceptual clustering

Abstract

In this paper, we consider unsupervised clustering as a combinatorial optimization problem. We focus on the use of Local Search procedures to optimize an association coefficient whose aim is to construct a couple of conceptual partitions, one on the set of objects and the other one on the set of attribute-value pairs. We present a study of the variation of the function in order to decrease the complexity of local search and to propose stochastic local search. Performances of the given algorithms are tested on synthetic data sets and the real data set Vote taken from the UCI Irvine repository. © Springer-Verlag Berlin Heidelberg 2001.