Community Detection by Affinity Propagation

Community structure in networks indicates groups of vertices within which are dense connections and between which are sparse connections. Community detection, an important topic in data mining and social network analysis, has attracted considerable research interests in recent years. Motivated by the idea that community detection is in fact a clustering problem on graphs, we propose several similarity metrics of vertex to transform a community detection problem into a clustering problem, and further adopt a recently-proposed clustering method, namely ‘Affinity Propagation’, to extract communities from graphs. We demonstrate that the method achieves significant quality in detecting community structures in both computer-generated and real-world network data in near-linear time. Furthermore, the method could automatically determine the number of communities.