Large-scale community detection based on a new dissimilarity measure

The process of the community reveal has become such a major problem in different fields, in particular for the social networks that have seen a growth over the last decade . Diverse ways of several methods have been proposed to resolve this inference problem. Nevertheless, the computational run time and the used space become a real handicap especially, when the networks are large. In this paper, we introduce a new approach of community detection that relies greatly on a new dissimilarity measure which allows to find the edges having more tendency to be between two communities. Then, we suppress them to have some preliminary communities. After that, we merge the induced subgraphs without using the modularity optimization to avoid its resolution limit. Finally, we evaluate our proposed method on real and artificial networks. The experiments show that our way of detecting communities outperforms or as effective as the existing algorithms (CNM, WalkTrap, InfoMap) while the used space and time complexity are better.