Community Detection Research Based On Optimizing High-Influence Vertexs In Social Network

With the development of social network, community detection has been a hot research topic. Study on the community helps explain the characteristics of network structure, which provides powerful support for information recommendation, information diffusion and many other applications. In recent years, the community detection techniques and methods have been well studied both in the academica and industrial community. However, as the increasing of social network size and the evolving of network structure, the lack of standardized definitions of communities and verification with ground-truth network communities, has restricted further research. Moreover, the experimental statistical analysis shows there are significant differences between results of traditional community detection algorithms and ground-truth communities, which motivated our research.In this paper we investigate the effects of nodes with high influence on community detection results of the network from two aspects as folliows.(1) We design a community detection framework based on semi-supervised machine learning, and proposed a high-degree-optimization lable propagation algorithm. Experimental results on several real-world datasets show the effects of nodes with high degree on the community detection results. Then we discussed several methods to differentiate whether the high-degree nodes can be optimized or not.(2) A variable influence community detection algorithm based on PageRank, as well as a variable influence local community detection algorithm from one seed vertex is proposed, which can adjust the size of the community according to the actual application scenarios and users’expectation. Also, they are efficient with low complexity and avoid the disadvantages of traditional label propagation algorithm such as monster communities.Finally, this paper summarizes the above work, and points out what can be done in future research.