Multiscale Community Detection Using a Label Propagation-Based Clustering Method in Complex Networks

Multiscale community detection algorithms can reveal the hierarchy of complex networks. However, the existing algorithms are unable to realize full-resolution community detection, and the hierarchy structure of the obtained community is overidealized. Aiming at these problems, we propose an algorithm named label propagation algorithm with multiscale community detection (LPAMCD), which introduces a two-phase propagation process and a tunable parameter, called the belonging coefficient threshold, into the label propagation algorithm to realize full-resolution community detection. Moreover, LPAMCD has the ability to find the mechanism of dynamic confrontation between adjacent communities in absorbing boundary nodes, which implies that the community hierarchy of social networks is not an idealized dendrogram. The extensive experimental results with real networks show that LPAMCD can detect community structures at full resolution scales with high accuracy and stability. Furthermore, the novel finding of dynamic confrontation is demonstrated in the experiments.