Research Of The Fuzzy And Evolution Clustering Algorithm In Social Network

Along with Facebook, Twitter, RenRen, QQ community, Sina microblogging and other social networking service platform successful promotion,the research in social networks is becoming increasingly important and widespread. Community structure is a common characteristic of these social networks, so-called community is the network of "group", closely linked group, contact group sparse. Traditional community discovery algorithms almost only can found non-overlapping community structures in a static network, but real-world social networks are often evolve over time and community structures usually can overlap..In this paper, we research the fuzzy clustering algorithm and evolutionary Clustering algorithm based on time series in social networks, thus completing the overlapping and dynamic communities’ detection.The accuracy of selected initial point in clustering has a very important impact on the efficiency and quality of cluster.To find the correct initial core point in front of the social network clustering, Based on the basic theory in Social Network, the theory of the strong and weak links and structural hole theory, we propose two algorithms to get the initial core points in social network clusting.we call these algorithm SH_SW_IP and SH_SW_DP, the algorithm considering the importance of nodes and node distance to obtain the initial core points, experimental results show that the algorithm can obtained good initial points within lower the time complexity and can give the approximate number of communities in a number of communities is unknown.Overlapping community’s detection is the most recent research focus, FCM algorithm is a classical fuzzy clustering algorithm. So we extends the theory of strong and weak links and with reference to six degrees of separation theory to constructs a node similarity defined methods, combined with FCM algorithm and the use of a stable way for the initial core point determination, then re-design the program to Get Local optimum in FCM algorithm within social networks, according to certain criteria and then set thresholds for each node to determine the class standard, which can also find structure of overlapping communities, the paper said the algorithm is SCCFCM algorithm, experimental results show that with the increase of data sets, SCCFCM algorithm shows better robustness.Dynamic communities detection is the latest hot spot and evolutionary clustering algorithm is one of its main research direction, but forgetting factor is difficult to determinate.this paper puts forward the concept of inertia node, pointing out the rule of the key node inertia variation, by comparing the Social networks of key nodes between different time, we get an approximate range of forgetting factor. After forgetting factor determined,we use evolutionary framework to improved SCCFCM clustering algorithm, so that it can dynamically discover overlapping we call the this algorithm ESCCFCM.Comparative experimental results show that the community detected by ESCCFCM not only has a high degree of Q but also exhibit better smoothness over different time steps.