Overlapping Community Discovery Algorithm Based on Label Propagation

Abstract

Based on the label propagation algorithm, the SLPA discovers the overlapping communities in the network through the dynamic process of interaction between Speaker and Listener. The time complexity is approximately linear. However, there is randomness in the process of label propagation, and the initialization of node labels takes a lot of resources when it is applied to large-scale networks. To solve the above problems, an overlapping community discovery algorithm LP-OCD based on label propagation is designed by improving SLPA. The algorithm pre-processes the network with the K-shell decomposition algorithm to remove the edge layer nodes before each node memory initializes the label. In the label propagation phase, the randomness of the algorithm is reduced by improving Speaking and Listening strategies. Labels of all edge layer nodes in the post-processing phase are determined by the information of their neighbors. Experimental results on social networks and synthetic networks show that the LP-OCD algorithm not only has approximately linear time complexity, but also significantly improves the quality of the overlapping communities discovered.