Inferring Community Members in Social Networks by Closeness Centrality Examination

Abstract

It is important task to discover communities or hidden groups by analyzing the messages collected in social networks. For the case when some members of a community are known, a proper method is still necessary to infer the remaining community members. To address such an issue, we develop a closeness centrality examination algorithm to obtain the remaining community members with some known community members. In the proposed model, the message connections among all social network members is captured by a weighted graph model where the edges are assigned with weights derived from the sensitivity of topics contained in the messages by text analysis. In addition, the nodes of known community members form a central sub tree in the weighted graph model. The suspicious priority list of possible community members is obtained by calculating a closeness centrality score to the central sub tree. With the priority list, the remaining community members can be determined using cluster analysis and outlier analysis. The proposed method is validated with experiments.