A novel granular approach for detecting dynamic online communities in social network

The great surge in the research of community discovery in complex network is going on due to its challenging aspects. Dynamicity and overlapping nature are among the common characteristics of these networks which are the main focus of this paper. In this research, we attempt to approximate the granular human-inspired viewpoints of the networks. This is especially helpful when making decisions with partial knowledge. In line with the principle of granular computing, in which precision is avoided, we define the micro- and macrogranules in two levels of nodes and communities, respectively. The proposed algorithm takes microgranules as input and outputs meaningful communities in rough macrocommunity form. For this purpose, the microgranules are drawn toward each other based on a new rough similarity measure defined in this paper. As a result, the structure of communities is revealed and adapted over time, according to the interactions observed in the network, and the number of communities is extracted automatically. The proposed model can deal with both the low and the sharp changes in the network. The algorithm is evaluated in multiple dynamic datasets and the results confirm the superiority of the proposed algorithm in various measures and scenarios.