Overlapping community detection in social networks with Voronoi and tolerance neighborhood-based method

Abstract

Community detection is typically viewed as a graph clustering problem with early detection algorithms focused on detecting non-overlapping communities and formulating various measures and optimization methods to evaluate the quality of clustering. In recent years, overlapping community detection especially in real-world social networks, has become an important and challenging research area since it introduces the possibility of membership of a vertex in more that one community. Overlapping community detection by its definition implies soft clustering and leads to an ideal application of granular computing methods. In this paper, a hybrid computational geometry approach with Voronoi diagrams and tolerance-based neighborhoods (VTNM) is used to detect overlapping communities in social networks. Voronoi partitioning results in a crisp partition of an Euclidean space and a tolerance relation makes it possible to obtain soft partitions. A Voronoi diagram is a method to partition a plane into regions based on nearness to points in a specific set of sites (seeds). In the VTNM approach, these seeds are used as cores for determining tolerance neighborhoods via a non-transitive binary relation. The intersection of these neighborhoods are used to discover overlapping communities. Our proposed VTNM algorithm was tested with 7 small real-world networks and compared with 11 well-known algorithms. VTNM algorithm shows promising results in terms of the Extended Modularity measure, Average F1-score and Normalized Mutual Information (NMI) measure.