Research On Key Technologies Of Overlapping Community Mining In Large Scale Social Networks

With the rapid development of information technology,social media emerging,people's production and life increasingly inseparable from the network.Because people product information on the social network,a large amount of network data has been produced.Social networks have tremendous scale,information-rich and dynamic characteristics,network data scientists are facing explosive growth,how to dig out the community from an increasingly complex social networks and use network data to guide the production of other issues is particularly important.Among the many characteristics of social networks,community structure is one of its unique concepts of structures.There are two types of communities:non-overlapping and overlapping community.In non-overlapping communities,a node can only belongs to one community.In overlapping communities,a node can belongs to many communities.Overlapping communities are more able to accurately reflect real-world network structure mining,and overlapping community for understanding the function of the network and predicting network behavior are more practical significance.However,the current common overlapping community discovery algorithm in the face of big data environment,the emergence of high complexity or community divided low accuracy problems.This paper uses social network as the object of study,based on a complex network theory,using mathematical modeling,algorithm design methods,experimental data and other methods to research the community mining of social networks.The main work is as follows:(1)The use of optimization algorithms to solve the problems of discovery overlapping community is an important idea.However,because of the module function's own limitations,they cannot guarantee the quality of community algorithm. Based on the existed Community structure evaluation index,this paper selects the community fitness and the minimum and maximum partition value of community as objective function,using multipurpose optimal method to mine social networks.The community mined by the algorithm can not only satisfy the tight connection features in communities,but also can satisfy the sparse connection features in communities.In order to mine more realistic overlapping communities,we select density and network variation,which fit overlapping communities,as objective functions.This paper proposes a multiobjective optimization of overlapping discovery algorithm,which reflects that overlapping communities have the high degree of modularity.(2)Due to the instability of the label propagation algorithm,propagation algorithm is used to directly label overlapping communities found that mining results will be quite different for each community and other issues.In order to improve the stableness,this paper proposes a stable overlapping community detection algorithm based on label propagation.Through setting the order of label propagation,changing the update strategy of labels,this algorithm improves the stableness and in the meanwhile,the complexity of algorithm is low.In order to further use overlapping community mining for weighted network,using the influence factors between the nodes as the basis for the label propagation,we designed an overlapping community algorithm which is for weighted network based on label propagation.This algorithm can improve the stableness of label propagation algorithm and is also suitable for weighted network.(3)In order to find a community mining algorithm which is suitable for both weighted network and unweighted network,the physics of gravitation principle is introduced into overlapping communities discovery and proposes a network of community-based weighting factor gravitation discovery algorithm.Firstly,the concept of gravity factor is given.We can detennine the strength of the connection between nodes and nodes and communities based on the size of gravity factors,and then determine which communities the nodes belong to.In the algorithm,the evaluation index of the weighted network is evaluated,and the characteristics of the network and the weighted network are considered.(4)A converged network node content and network topology discovery algorithm overlapping communities.According to the characteristics of the microblog network,the content information of the nodes in the microblog network and the relationship between the users in the microblog network are fused.The RT model of the micro blog network is established,and the micro blog network is mapped to the weighted network.In order to make full use of the semantic information of micro blog network,this paper proposes static model and dynamic model of topic mining in micro blog network,which can dig out the interests of the authors in the micro blog,and then calculate the similarity between the interests of the authors in the micro blog network.Based on this,we use label propagation algorithm for overlapping community mining based on micro blog network.Experimental results show that the proposed algorithm is more semantic cohesion.In general,the study of this paper,has a significant influence on improving the quality of community mining,reducing the complexity of community mining algorithm,and solving the problem of low quality of community caused by the data sparsity of social network.