ma-CODE: A multi-phase approach on community detection in evolving networks

Abstract

Abstract Detecting communities or clusters in networks becomes a decisive issue in various interdisciplinary areas in recent years. Numerous methods are proposed to uncover community in networks, although the fundamental problem of most of the methods is the evolving nature of the networks and the presence of the imprecise number of communities. Since, real-world networks are scale-free networks and due to the preferential attachment properties, the low degree nodes are attracted towards the hub nodes showing the power-law distributions. Hub nodes are highly surrounded by their neighbors and connectedness among the nodes within a community is larger than the others. As a result, the underlying structural details facilitate to uncover precise community structure. In this work, we present a multi-phase model ma-CODE to uncover communities based on the inherent association without having any prior information about the presence of the number of communities. The multi-phase approach contains the identification of high degree nodes, label propagation and community merging. The high degree nodes are identified based on the voting by the adjacent members; the label propagation is to assign the same community identification number to those members showing high similarity; the community merging is performed among the different communities only when there is a significant increase in the modularity after combination. We examine the competence of our proposed methods in the light of twelve (12) popular real-world social networks and eight (08) artificial networks. Experiments and simulation results using five (05) different statistical assessment parameters show that ma-CODE is superior over contemporary community detection methods.