Overlapping Community Detection Using NVPA

In this paper we propose two algorithms for overlapping community detection based on neighborhood vector propagation algorithm(NVPA), a community detection algorithm which can detect disjoint communities with high accuracy. The first algorithm is named Link Partition of Overlapping Communities (LPOC). In this algorithm, we first convert a node graph to a link graph, then we use NVPA to find the communities on the link graph. After converting link communities back to node communities, overlapping nodes are filtered to decide whether they belong to multiple clusters or not. This algorithm retains the high accuracy of NVPA, and overcomes the drawback of link clustering which typically produces too many overlapping vertices. The LPOC algorithm relies on link graph clustering which has a high computing complexity. To overcome this, we propose another algorithm named candidate overlapping nodes screening( CONS) algorithm, which uses NVPA on node graph to find non-overlapping communities, then we design a quality function and a screening method to identify nodes that really connect multiple communities. We evaluated these two algorithms on both LFR benchmarks and real-world networks, and compare with several other approaches. Results show that our algorithms have high accuracy, and perform much better than many other algorithms in terms of partition density, modularity and WAC value.