An Empirical Study of Community Overlap: Ground-truth, Algorithmic Solutions, and Implications

Abstract

In real-world social networks, communities tend to be overlapped with each other because a vertex can belong to multiple communities. To identify these overlapping communities, a number of overlapping community detection methods have been proposed over the recent years. However, there have been very few studies on the characteristics and the implications of the community overlap. In this paper, we investigate the properties of the nodes and the edges placed within the overlapped regions between the communities using the ground-truth communities as well as algorithmic communities derived from the state-of-the-art overlapping community detection methods. We find that the overlapped nodes and the overlapped edges play different roles from the ones that are not in the overlapped regions. Using real-world data, we empirically show that the highly overlapped nodes are involved in structure holes of a network. Also, we show that the overlapped nodes and edges play an important role in forming new links in evolving networks and diffusing information through a network.