Research On Algorithms For Community Detection In Social Networks

Community detection is a very important job in social network analysis.Through community detection,it can discover the organizational structure information and the social function of the network.Based on the rich and different views data in social network,the fusion of multi-view learning and different types of community detection methods are studied to improve the detection accuracy.The main research work is as follows:1.The follow,mention and retweet relations between social network users and the prior information are fused into the community detection based on genetic algorithm,a community detection method is proposed by combing multi-view information,prior information and genetic algoritm.The network modularity is taken as the optimization objective function,the user's follow,mention and retweet relations are considered and combined with genetic algorithm for network community detection.Furthermore,the prior information for the community division of users is also fused into the detection algorithm to guide the evolutionary search process.The simulation results show that the fusion of multi-view information and prior information can improve the accuracy of network community detection.2.The selective ensemble learning method is applied to the community detection based on genetic algorithm,and a method of community detection based on selective ensemble is proposed.For the fluctuations of the detection results obtained by genetic algorithm,the ensemble of multiple results of network community detection is studie and two kinds of selective ensemble method is presented.The first selective ensemble method is based on genetic optimization,the other is based on Pareto evolution optimization.For the proposed community detection method,the multiple network community detection results are obtained by genetic algorithm and multi-view information.Then,the multiple network community division results are ensembled by using evolutionary optimization algorithm and the final detection results are got.The simulation results show that selective ensemble can improve the accuracy of network community detection.3.Aiming at the fusion problem of linking relational data and content data in network community detection,a method of community detection based on multi-link relation and content attribute is proposed.First,the different netwok links are fused and the false information is eliminated.Then,the graph model is constructed by using nearest neighbor structure of network links.Next,the nearest neighbor graph and the content of data are fused by using a symmetric NMF,the constraints of ownership matrix is relaxed by the introduction of the difference function between the different ownership matrixs.Last,the effective iterative method is designed to obtain more accurate results of community division.The experimental results show that theproposed algorithm can fuse the two kinds of information with different nature effectively and the community detection result is more real.It is useful for different view data with big quality difference and can maintain the stability of the results.4.Aiming at the problem of missing data of some users' nodes in network community detection,this paper studies the multi-view heterogeneous community detection problem under the condition of missing data,and proposes a fusion community detection method with some node data missing.This method constructs two regular items dealing with missing data,one is that the user who does not participate in the fusion is not involved in the fusion when comparing the different view data;the other is the nearest neighbor data using the other view of the missing data user as a substitute for information,to participate in the integration of perspective.On this basis,a heterogeneous multi-view community detection algorithm is proposed based on two regular terms.The experimental results on the real data set show that the proposed algorithm can alleviate the problem of community detection with large difference in different views and lack of data,and obtain real and reliable community detection results.