Research On MMSB Based Community Detection Algorithm In Weighted Social Network

Community structure is an important feature of social networking. It is helpful to the deep understanding of the network topology characteristics in macro, exploring the network user’s behavior characteristics and the logical relationship between the usersand the transmission characteristics of the information in micro.Therefore, to find the community structure of social networks has become particularly important.Most of the existing social network community discovery methods can only be entitled to the powerless or undirected network to find overlapping communities, then appeared mixed membership random block model(MMSB) community detection method.MMSB would not only find overlapping communities in powerless and directed network, but also the quantitative membership degree of node in each community. However,MMSB does not apply to the weighted network, therefore, this paperproposes a method to find the overlapping community in directed weighted network based on MMSB.The article’s main work is as follows:1) Proposing an overlapping community detection method which is called weighted mixed membership stochastic block model(WMMSB) of directed weighted networkon the basis of MMSB.This method firstly by establishing a statistical model to simulate the observation network, then estimate parameters in the model using the rule of maximum likelihood, according to mixed membership degreeof the node, we can find the network community.As the likelihood function of the parameter is extremely complex, the traditional maximum likelihood estimation method cannot get the estimates of parameters, so the article uses the variational expectation maximization(VEM) algorithm to estimate parameters.2) The article proposed splitting result judgment criteria of the nodes in view of the mixed membership degree. This method firstly obtained the edge number of node in everycommunity by mixed membership and then judge the community by edge number and the mixed membership degree, if they select the same community,which means the node is divided in the correct community, otherwise the node is divided in the wrong community.3) Applies WMMSB to journal reference network, display and analyzeresult of community detection. Then explain WMMSB can correctly divide community, and can quantitatively get the advantage of intimacy between nodes and the communitythrough the contrast tothe community detection method that based on the edge direction.4) Using the web crawler technology to get the data of user in Sina microblog, and divide the network respectively with MMSB and WMMSB, and display the network node distribution by using the method of visualization.The simulation results showed that firstly the logarithmic likelihood function of the parameters in WMMSB can converge; Second, WMMSB can find out the optimum number of community by maximum likelihood function;Then, we can divide the network by mixed membership value of each node;Finally, by comparing WMMSB and MMSB corporate division results, combining with the visual network diagram, found WMMSB is better than MMSB in node classification accuracy rate and module quantity value of community structure.