Research Of Overlapping Community Detection Algorithms Combined With Influence

With the rapid development of the Internet technology, online social networks become larger and larger. The complex and overlapping structure these networks present makes community detection a problem. Another feature of social networks is information propagation characteristics, and it arouses a social impact maximization problem. Currently, the overlapping community detection algorithms mainly focus on the network structure to divide the community, the results of which need an improvement in quality. In this paper, we take the information propagation and social impact maximization into considerations, improving the accuracy of the existing label propagation overlapping community detection methods significantly through combining label influence factor both in label initialization and the propagation process.In this paper, we do a deep research in both social impact maximization area and community detection area, and then propose a label propagation algorithm for overlapping community detection algorithm combined with influence. We find the most influential k nodes of network through a parameter-adjusted social influence maximization algorithm, and extend these nodes to maximal cliques as an initial unit for label propagation. Then we propose a method to compute the influence of label globally to get the belonging coefficient of the nodes in the cliques, so that the overlapping nodes can be assigned to some labels correctly in percentage. In label propagation process, the influence of labels should also be taken into consideration and the order of propagation should be reasonably set. Experiments on four networks of different structure generated by LFR benchmarks and on a real public dataset show that in the case of graphs of relatively small community size, our algorithm has a significant improvement in community quality than the existing label propagation overlapping community detection algorithms.