Bidirectional Label Propagation for Community Detection

In complex networks, communities often show the presence of homophily between members, since homophily is the tendency of individuals to associate and bound with similar others to form densely interconnected groups. In this paper, we propose a new bidirectional label propagation community detection algorithm, called Dis-Sim, based on the fundamental idea that a node and its most similar neighbors belong to the same community, while a node and its most dissimilar neighbors belong to different communities. The proposed algorithm depends on the network structure alone and explores the dissimilarity and similarity between each pair of nodes to identify the communities. The dissimilarity measure will separate nodes into different groups, while the similarity measure will merge nodes into clusters. The empirical experiments on real-world networks against state-of-the-art algorithms are used to evaluate the performance of the proposed method. The results confirm its effectiveness and computational efficiency.