Community Detection Using Label Propagation in Entropic Order

Community structure is an important topological property of complex networks and community detection has become a common methodology to understand the organization and function of various complex networks. Label propagation is an extremely fast algorithm with near linear time complexity and it is applicable to large-scale networks. In spite of the advantages of label propagation, the issue of the robustness of the algorithm has not yet been properly addressed. Random node updating orders may result in totally different network partitions and even unreasonable community structures. In this paper, we define a new measure which is called label entropy by introducing the information theory. And we also propose label propagation in entropic order whose main idea is updating the labels in the order from small to large by label entropy. We test our algorithm on both computer-generated networks and real-world networks. The experimental results demonstrate that label propagation in entropic order not only is more robust than label propagation, but also improves the performance of community detection. Moreover, label propagation in entropic order doesn´t increase the time complexity and it retains the algorithmic simplicity of label propagation.