Disjoint Community Detection in Networks Based on the Relative Association of Members

We propose in this paper a hybrid system called DCDRAM that detects disjoint communities. It is based, in part, on the underlying techniques of the network-centric, hierarchycentric, vertex-centric, and group-centric approaches. It adopts most of the underlying techniques of the four approaches. Most of these approaches work well in only networks with certain topologies. DCDRAM aims at overcoming the limitations of each of the four approaches to enable it to work well in networks with all types of topologies. It does so by: 1) measuring the betweenness of each edge (u, v) in such a way that the betweenness acts as an indicator of the influences of vertices u and v over the flow of information in the entire network; 2) employing a novel logarithm-based formula that captures and further enhances the eigenvector principle in order to characterize the global influence of each vertex in the network; 3) employing a novel agglomerative-like formula that discovers natural divisions of a network; and 4) employing a novel belonging formula that helps in discovering disjoint communities. We evaluated DCDRAM by comparing it empirically and experimentally with nine methods. Results showed marked improvement.