Based On Density Modules Microblogging Community Discovery

Microblogs have created a fast growing social network on the Internet and community detection is one of the subjects of great interest microblog network analysis. The "follow" relationships highlighted in many existing methods are not sufficient to capture the actual relationship strength between users, which leads to imprecise communities of microblog-users. In addition, because many popular community detection algorithms fail to work with non-supervision, low time complexity or high resolution ratio, it has become another factor affecting the quality of community partition.Including the microblog network, the social network analysis originally belongs to the research categories of sociology and physics. The researches of community detection of sociology and physics have gained richer fruits than that of computer science. So, solving existing communities detection problems with the help of the advanced knowledge in sociology and physics is a good way of thinking, and already become commonly method accepted by social network researchers in the various domains. Sociologists believe that the strength of the relationship is often indicated by the frequency and reciprocity of interactions:the higher frequency and reciprocity interactions occur between a pair of users, their relationship is closer. According to this view, we can find a way to quantify the microblog-user relationship strength for enhancing the accuracy of community detection from the perspective of model. In addition, the idea of hierarchical clustering proposed by sociologist does not require a priori knowledge acquisition, which gives us a good way to design an algorithm with non-supervision; Physics puts forward the concept of modularity density can accurately depict small community structure, which can help to solve the problem of resolution limit.More specifically, the main work and contributions can be summarized as follows:(1) User relationship strength of graph modelWe select the mutual and frequency of interaction such as "forward","comment" as the basis of modeling. According to the sociological theory, we proposed a new metric based on the interaction activities to obtain the relationship strength quantitatively, then we presented the graph-based method for modeling the microblog-users relationship.(2) The hierarchical clustering algorithm based on modularity densityTo discover the communities precisely, we proposed the concept of the subgraph proximity and gave a hierarchical algorithm by incorporating the quantitative relationship strength and the modularity density criterion. By this algorithm, we can discover the communities precisely with high resolution ratio.(3) The experimental resultsBy applying our algorithm to analyzing some real social networks and artificial networks, we tested and verified the effectiveness, applicability and efficiency of our algorithm. By comparing to the existing algorithm, we demonstrated the advantages and disadvantages of our algorithm.