Research On The Problem Of Community Discovery In Social Networks

Recently,with the rise of social networks,community detection research in the area of data mining and social network analysis has been discussed by more and more pople,which can effectively explore the community structure hidden in social networks.Community structure,which reflects the characteristics of user behavior in a network and the connection strength between them,is crucial to understanding the structure of the entire network and analyzing the user interactions.At present,research on community detection has produced many results,and many community detection algorithms have been designed.However,issues are still existed.For instance,further research is essential about the relationship between nodes and the large dimension of node attributes with deep structural information.The higher precision is still necessary.According to the consisting difficulties of community detection,this paper is focus on the area of community detection in social networks,and proposes three improved community detection algorithms to solve the above problems from different angles.The specific work can be briefly summarized as the following parts:The first part is mainly about the limitation of modularity function.There are questions to be raised about a community detection algorithm CDPSO based on multi-objective.The intra-community density NRA and the inter-community density RC are used to model the multi-objectives,and a selection strategy is used to select the optimal solution set.The algorithm has clear advantages in community detection quality and time complexity.The second part is mainly about the problems of insufficient research on the relationship between nodes and unreasonable initial selection of community centers.A community detection algorithm CDP-EW based on edge influence weight is proposed.By using the degree centrality of the node to calculate the influence of the node and make the initial community center selection.For the link relationship between different nodes,the edge influence weight is redefined to calculate the similarity between nodes.The algorithm can efficiently mine overlapping communities in the network and shows excellent properties in terms of modularity.The third part is mainly about the problems of large node dimensions in real networks,difficult to find important information,and difficult to integrate deep structural information to effectively divide communities.A community detection algorithm GMCD based on graph neural network embedding is proposed.The proposed method utilizes Graph Neural Network(GNN)and Gaussian Mixture Distribution(GMM),learns node embedding through graph neural network to obtain node representation vector,and inputs the node representation vector into Gaussian Mixture Distribution model for clustering operation.The algorithm has good network representation performance,all evaluation indicators have been improved,and the effect of community detection is good.