Research On Community Detection And Influence Analysis In Social Networks

With the rapid development of computer technology, new-style social networks haveemerged continuously, such as Sina micro-blog, Renren, Facebook and Twitter ect. The socialnetwork service has timeliness and dynamic features with the help of mobile communication.And it has satisfied people to share and exchange information with others in anytime andanywhere. But there are many safe hidden and public opinion transmission problems with thebringing of convenience and flexibility at the same time. The research of community structurediscovery and influence propagation analysis technology could help us to understand theessential features and inner structure of network more deeply, and provide decision support fornetwork attack of information and behavior layers. And now there are three important problemsto be solved about the two aspects.(1) Most static community detecting algorithms are usuallybased on the structure characteristics of network and lack of consideration of social attributeinformation.(2) The research on static network is not enough to describe dynamic property andessential structure of real networks.(3) The large time cost of existing influence algorithms is notfit in with the social networks that the scale becomes more and more large.To solve these problems, this paper researches on community structure discoverytechnology supported by the project of the national863high technology development plan. First,a community detection algorithm was proposed that combining structure and attribute ofnetworks. And based on it dynamic communities by the change of network topology with timechanging were researched. Finally, the problem of social network influence propagation wasanalyzed based on better community structures. The main work and achievements are as follows.1. A community detection algorithm is proposed based on fuzzy equivalence relationcombining structure and attribute in social networks. In this approach, a new concept ofintegrated dissimilarity distance index is used for combining structure and attribute to scalesynthesis distance between nodes, and regarded it as the subordinate relation to build the fuzzyequivalence relation matrix. Then appropriate clustering threshold value was choosen to used forcommunity detection. The applications in the real social networks demonstrate that it coulddetect denser networks better. The higher accuracy rate is showed and the vertices in the samecommunity are homogeneous and connected close.2. An incremental dynamic community detecting algorithm is proposed based on attributeweighted networks. Based on static community structure, the current community structure isupdated by the incremental comparing with previous time one. The dynamic nodes’ communityis dicided by the concept of topological potential attraction between nodes and communities. Theexperiment on real network data proved that the algorithm was more effectively and timely todiscover meaningful community structure with a smaller time complexity.3. An influence evaluation model is proposed based on network community structure forlarge time cost. Firstly nodes influence in each community is evaluated and core members aredigged. To find a small subset of nodes in the set consisted by the core nodes and linkingcommunity nodes to get the maximization diffusion with minimization cost. Experimental result demonstrates that the model achieves the subset with more abroad influence diffusion andreduces running time compared with traditional methods.