Study Of Community Detection Algorithm In Signed Networks

There are many relationships among entities in social networks at present,such as cooperation and competition,friends and enemies,support and opposition,etc.The social networks with positive and negative relationships are called signed networks.As a special case of social networks,signed networks have become one of the hotest research topics for domestic and foreign scholars.The study on community detection in signed networks is the basis of network analysis,which has important influence on the prediction,personalized recommendation,and user characteristic analysis and so on.Nowadays,the research on community detection in signed networks is mainly divided into two types,the method based on optimizing objective function and two-phase method based on heuristic.Aiming at the problems of existing methods,this paper makes the following research in order to improve the accuracy of community detection and the stability of the community.Firstly,as the classical two-phase signed networks community detection algorithms,CRA can not correctly divide the network because the negative edges are ignored.The TFCRA algorithm is proposed as a new two-phase fusion signed networks community detection.To improve the accuracy of community detection,the algorithm defines the assignment rules of vertices with negative edge and converge the two phases through taking into account the positive and negative edges at the same time.The correctness and rationality of the TFCRA algorithm are verified by experiments.Secondly,aiming at the problem of poor stability in current algorithms,and based on the idea that the more common neighbors the higher similarity between vertices,we extend similarity in the traditional social network to signed networks and propose a new BNS_SNCD algorithm for the community detection in signed networks based on similarity.The algorithm represents the similarity with the number of common neighbors and divides the community based on similarity degree.And it improves the efficiency of community merger by the commbining strategy from small community to big community.Finally,the correctness and rationality of the BNS_SNCD algorithm are verified by experiments.Thirdly,in view of the problem that the current algorithms only consider the number of common neighbors,a new signed network community detection algorithm BTCN_SNCD is proposed based on the tightness of common neighbors.The algorithm takes the contribution degree,tightness among nodes and communities,overlap coefficient and community tightness as criterion,and can find more compact initial communities,and it also can merge the overlapping community.This ensures the accuracy of the community detection.The correctness and efficiency of the BTNC_SNCD algorithm are verified by experiments.Fourthly,for the local oscillation problem caused by determining the assignment of nodes with negative edge in current algorithms,a algorithm SBTNS_SNCD is proposed based on structural balance theory.Combining with the positive and negative degree,we propose a new similarity and participation degree calculation method,which can reduce the oscillation and obtain more accurate community structure.The correctness and efficiency of the SBTNS_SNCD algorithm are verified by experiments.Finally,based on the SBTNS_SNCD algorithm,we put forward the concept of core nodes and shell nodes.Combining with the idea that the shell nodes are closely related to other communities and may be overlapping nodes,the signed networks overlapping community detection algorithm SNOCD is proposed.It can accurately find the overlapping nodes,and then detect the signed networks overlapping community.Compared with other algorithms,SNOCD can be more accurate and flexible to detect the signed networks overlapping community.