A New Encoding Method for Graph Clustering Problem

Abstract

Clustering is used as an important technique to extract patterns from big data in various fields. Graph clustering as a subset of clustering has a lot of practical applications. Due to the NP-hardness of the graph clustering problem, many evolutionary algorithms, particularly the genetic algorithm have been presented. One of the most effective operators on the performance of the genetic algorithm is how to represent the solutions of a problem (i.e. encoding). The number of possible partitions of a graph is equal to Bell Number. In the literature, three encoding methods have been presented for graph clustering problem. The number of partitions that these encodings can generate is more than the Bell Number; which indicates that these methods generate a large number of same and iterative solutions which makes the speed of obtaining the solution unacceptable and leads to this fact that the good space search encounters a problem. To overcome this drawback, in this paper we present a new encoding method for graph clustering problem where the number of the generated solutions by this encoding is exactly equal to the Bell numbers. The initial results of our experiments represent that the quality of the obtained solutions by the new encoding is promising.