Abstract
In this paper, we propose an algorithm that detects overlapping communities in networks (graphs) based on a simple node behavior model. The key idea in the proposed algorithm is to find communities in an agglomerative manner such that every detected community S has the following property: For each node i ε S, we have (i) the fraction of nodes in S / {i} that are neighbors of node i is greater than a given threshold, or (ii) the fraction of neighbors of node i that are in S / {i} is greater than another given threshold. Through computer simulations of random graphs with built-in overlapping community structure, including LFR benchmark random graphs and Erdös-Rényi type random graphs, we show that our algorithm has excellent performance. Furthermore, we apply our algorithm to several real-world networks and show that the overlapping communities detected by our algorithm are very close to the known communities in these networks. © 2013 IEEE.