A Method Of Community Discovery And Dynamic Network Evolution Analysis Based On Similarity

Social network as the carrier of large and complex information,is gradually becoming the hot research field,a number of community discovery algorithms come out,but in the practical application,every algorithm is facing the situation of low quality,low efficiency,small scope of trial test.The accurate and efficient algorithm has positive significance in the mining community,to personalized recommendation service and monitoring public opinion has far-reaching influence.In this paper,the research status and relevant theories of social networks and community discovery are introduced,and the advantages and disadvantages of the traditional community discovery algorithms are summarized,aiming at these shortages,an improved GN algorithm is proposed.First,by using the principal component analysis,user similarity is constructed based on the users' several attributes,and a weighted network is built up.Then,aiming at the problem of low efficiency of traditional GN algorithm,an idea of distributed computing is used to compute the edge betweenness,and in the calculation of edge betweenness,the weight of edges is joined.The improved GN algorithm can compute parallelly,save computing time and is suitable for the weighted networks.Second,this paper puts forward the concept of community contribution,makes the isolated nodes to find its community.Finally,by a careful study of changing rule of the community structure based on time series,three changing indexes of network are extracted,namely the number of nodes,the number of edges and community size,and an evolution model of dynamic network come up to rationally predict the existing network based on the changing rule of community structure.Finally,the paper use Matlab for data processing,to verify the improved GN algorithm.The experimental results show that the improved GN algorithm can not only guarantee the quality of divided community,but also reduce greatly computing time and isolated nodes.The experiment of several divided networks based on chronological order shows that the evolution model can achieve better predicting results,and maintain the stability of the community.