Research On Non-Overlapping And Overlapping Community Detection Algorithms Based On Core Nodes

Community detection is an unsupervised or semi supervised method based on graph mining.Its main purpose is to discover the high order similarity between nodes by exploring the topological structure of a graph,and classify these similar nodes into the same community.In societies,members are closely connected,while members of different societies have fewer connections.This method can be used in many applications such as user shopping recommendations and advertisements in e-commerce,protein and protein interaction prediction in bioinformatics,literature recommendation in citation analysis,or scholar collaboration.In order to explore more effective community detection in complex networks,this study focuses on the following issues:(1)High quality community detection in large and complex networks typically relies on the topology of graphs to partition node sets.However,real-world networks often have noisy and cluster independent links,which may lead to models dividing nodes from different clusters together;(2)Most community detection models based on graph neural networks still use K-means to locate community centers,and do not fully utilize the advantages of graph neural networks in learning node representations;The main work of the paper includes:First,a community detection algorithm(GRCD)for large-scale networks under graph reconstruction.This method can handle community detection in large-scale complex networks.First,delete the interconnected edges between communities to reconstruct the community structure of the original graph.GRCD then views the web as a social system,aiming to reveal communities in a more intuitive way;This paper proposes an efficient community detection strategy: a discourse power based community organization generation strategy.Finally,experiments are conducted on data sets of different scales.The experimental results show that the GRCD algorithm can not only handle large-scale networks,but also maintain high stability while maintaining a strong competitiveness in the quality of community division compared to several existing benchmark algorithms.Second,overlapping community detection(over DGI)of global and local mutual information for joint graphs.This is a graph neural network for dealing with overlapping community detection problems.Firstly,maximizing graph mutual information makes node vector representation more accurate in expressing graph structure information.Then,based on using the graph structure to locate the community center,the vector representation distance between nodes belonging to the same community is made closer and closer to the community center by maximizing the mutual information of the community.Finally,a target distribution is designed to help the model better solve overlapping community detection tasks.Through comprehensive experiments,it is shown that the performance of over DGI in overlapping community division is highly competitive compared to several existing benchmark algorithms.The main goal of this article is to explore how to use core nodes to improve the effectiveness of community detection.In GRCD,community organization strategies based on discourse power are used to focus on using core nodes to improve the accuracy of community detection.In over DGI,the vector representation distance between core nodes belonging to the same community is closer and closer to the center of the community by maximizing community mutual information.Both of these methods exhibit good performance,providing a new perspective for community structure mining in networks.