عنوان مقاله [English]
Detecting community structures is applicable in a wide range of scientific fields such as biological and social sciences. Community detection is one of the most renowned problems in the field of social networks mining. Thus, many methods have been introduced and developed in order to meet diverse needs of community detection. The aim of community detection is to partition the network
in such a way that relations between components of network are dense. Since the relations between the members of partitions are strong, it is possible to consider them as a community or a cluster. In this paper, we have considered community detection as a multi-objective problem. The objective functions are modularity and community scores, which are two of the most well-known
objectives in the literature. In order to optimize these objective functions, two algorithms, which are the enhanced versions of NSGAII and NRGA, have been proposed. These methods use a greedy algorithm to obtain initial population. Moreover, new crossover and mutation operators have been designed. The crossover operator is based on closeness of nodes. The mutation operator is based on TOPSIS method. The proposed crossover and mutation operators always generate feasible solutions. Furthermore, the closeness index helps to form distinct and high quality communities. We have compared the performance of the proposed methods with those of classical NSGAII, NRGA, and a well-known method called MOGA-Net by conducting several numerical experiments in six real-world networks. The experiments split into two parts. In the first part, we have compared the solutions of these five algorithms regarding the values of objective functions. The second part is dedicated to the comparisons made based on various multi-objective metrics. We have considered spacing, generational distance, inverted generational distance, set coverage, normalized mutual information, computation time, and the number of non-dominated solutions obtained by each method. In order to ensure that the solutions obtained by the proposed algorithms are significantly better than the ones provided by the other three methods, we have conducted several two-sample t-tests. The results showed significant improvement, and the proposed algorithms outperformed the other three methods regarding various criteria.