Multi-objective genetic algorithms, NSGA-II and SPEA2, for document clustering

Abstract

This paper proposes the multi-objective genetic algorithm (MOGA) for document clustering. The studied, hierarchical agglomerative algorithms,k-means algorithm and general genetic algorithm (GA) are more progressing in document clustering. However, in hierarchical agglomerative algorithms, efficiency is a problem (O(n 2logn)), k-means algorithm depends on too much the initial centroids, and general GA can converge to the local optimal value when defining an objective function which is not suitable. In this paper, two of MOGA's algorithms, NSGA-II and SPEA2 are applied to document clustering in order to complete these disadvantages. We compare to NSGA-II, SPEA2 and the existing clustering algorithms (k-means, general GA). Our experimental results show the average values of NSGA-II and SPEA2 are about 28% higher the clustering performance than the k-means algorithm and about 17% higher the clustering performance than the general GA. © 2011 Springer-Verlag.