A parameter-free community detection method based on centrality and dispersion of nodes in complex networks

•Parameter-free to decide initial centers and the number of communities.•A modified centrality measurement is presented.•High accuracy of the proposed K-rank-D compared with other algorithms.

K-means is a simple and efficient clustering algorithm to detect communities in networks. However, it may suffer from a bad choice of initial seeds (also called centers) that seriously affect the clustering accuracy and the convergence rate. Additionally, in K-means, the number of communities should be specified in advance. Till now, it is still an open problem on how to select initial seeds and how to determine the number of communities. In this study, a new parameter-free community detection method (named K-rank-D) was proposed. First, based on the fact that good initial seeds usually have high importance and are dispersedly located in a network, we proposed a modified PageRank centrality to evaluate the importance of a node, and drew a decision graph to depict the importance and the dispersion of nodes. Then, the initial seeds and the number of communities were selected from the decision graph actively and intuitively as the ‘start’ parameter of K-means. Experimental results on synthetic and real-world networks demonstrate the superior performance of our approach over competing methods for community detection.