An Improved Evolutionary Approach for Document Clustering

Abstract

Traditional approaches representing chromosomes based on cluster centroids do not allow dividing cluster centroids during crossover operations. Hence, significant diversity may not be achieved while the algorithm iterates from generation to generation. In our approach presented in this paper, we allow a crossover point to be even inside a cluster centroid, modifying some cluster centroids during the crossover operations. Such modifications also guide the algorithm to get rid of the local minima. Thus the proposed approach can find a better solution than the traditional approaches.