Detecting Disjoint Communities in a Social Network Based on the Degrees of Association between Edges and Influential Nodes

Detecting communities is crucial to understanding the dynamics of their members. However, the detection of 'good' communities is deemed demonstrably problematic, which is mainly due to the following two factors. First, real-world networks are complex and require optimizing multi-objective functions for capturing their community structures, whereas most current approaches optimize only one or two objective functions. Second, most current approaches detect communities in respect of the independence of how closely associated their connections are based on the global relative influences of the edges connecting them. To overcome these limitations, a clustering method needs to optimize multi-objective functions and employ global preprocessing techniques that consider the topology of the entire network. We, therefore, proposed a system called DAVE, which optimizes four objective functions that capture the community structures in most real-word network settings, and detects communities with regards of how closely associated their connections are based on the relative influences of the edges connecting them. We proposed novel formulas that capture these functions. Our method is the first to utilize the prediction of node-node associations based on global node-edge degrees of association. After ranking nodes based on their global relative influences on the network, some of the top-ranked ones will serve as core seeds for constructing communities. Then, the degrees of association between influential edges and seed nodes are computed. DAVE assigns a node to a community, only if each edge in the shortest path from this node to the community's core seed node is both influential and has significant degree of association with the core node. We evaluated DAVE by comparing it empirically and experimentally with 16 methods. Results showed a remarkable improvement.