Multi-objective clustering technique based on k-nodes update policy and similarity matrix for mining communities in social networks

This paper proposes a relatively all-purpose network clustering technique based on the framework of multi-objective evolutionary algorithms, which can effectively dispose the issue of community detection in unsigned social networks, as well as in signed social networks. Firstly, we formulate a generalized similarity function to construct a similarity matrix, and then a pre-partitioning strategy is projected according to the similarity matrix. The pre-partitioning strategy merely considers nodes with high similarity values, which avoids the interference of noise nodes during the label update phase. In this way, at the initial phase of the algorithm, nodes with strong connections are fleetly gathered into sub-communities. Secondly, we elaborately devise a crossover operator, called cross-merging operator, to merge sub-communities generated by the pre-partitioning technique. Moreover, a special mutation operator, based on the similarity matrix of nodes, is implemented to adjust boundary nodes connecting different communities. Finally, to handle different types of networks, we, therefore, have presented the novel multi-objective optimization models for this issue. Through a bulk of rigorous experiments on both unsigned and signed social networks, the preeminent clustering performance illustrate that the proposed algorithm is capable of mining communities effectively.