Overlapping community detection algorithm based on fuzzy hierarchical clustering in social network

A detection algorithm for overlapping communities based on fuzzy hierarchical clustering, CDHC, is proposed to detect the overlapping communities and to solve the fuzzy and hierarchical relationships among communities in social networks. The algorithm first utilizes the distance weighting factors to calculate the similarity among communities, and the communities with similarity larger than a given threshold are then merged together. The membership grade of each node for the merged community is computed and nodes with membership grades less than a given threshold are removed from the community to form a structure of the final overlapping community. The algorithm can not only detect the overlapping communities, but also detect the isolated nodes. The effectiveness of the proposed algorithm is tested through comparing it with two existing overlapping community detection algorithms, CMP and LFM, on the Lancichinetti synthetic network and real network datasets. Results show that the size of network and size of communities have little effect on accuracy of detecting communities, and the main factor to affect the accuracy is the mixed degree among communities. The detection accuracy of the CDHC on social networks with small communities is higher than that of LFM, and it is better than CMP on networks with large communities. The CDHC algorithm improves the detection accuracy while its stability is good. Therefore, it can be concluded that the CDHC is an effective overlapping community detection algorithm for social networks.