A Generalized Framework for Detecting Social Network Communities by the Scanning Method

摘要

With the popularity of social media, recognizing and analyzing social network patterns have become important issues. A society offers a wide variety of possible communities, such as schools, families, firms and many others. The study and detection of these communities have been popular among business and social science researchers. Under the Poisson random graph assumption, the scan statistics have been verified as a useful tool to determine the statistical significance of both structure and attribute clusters in networks. However, the Poisson random graph assumption may not be fulfilled in all networks. In this paper, we first generalize the scan statistics by considering the individual diversity of each edge. Then we construct the random connection probability model and the logit model, and demonstrate the effectiveness of the generalized method. Simulation studies show that the generalized method has better detection when compared to the existing methods.