Research Of Community Detection Algorithm Based On Average Mutual Information

With the rapid development of the Internet and Internet of Things technologies,Community detection in complex networks has become a research hotspot.Community Detection has very important practical value in user recommendation,public opinion monitoring,public safety,etc.How to evaluate the quality of the community detection algorithm is of great significance to the research and application of community detection.This paper focuses on community partition evaluation methods and community detection algorithms,including the following three aspects:1)In view of the defects of existing evaluation method,this paper proposes a new community partition evaluation method based on Average Mutual Information(AMI).The evaluation method uses the average mutual information value to measure the amount of information lost by community partition,and then measures the quality of community partition.Finally,the AMI method will be tested on real networks and artificial networks.Experimental results show that the AMI method can not only avoid the Resolution Limit problem existing in the modularity method,but also has a high degree of community partition judgment accuracy.2)In order to solve the problem of community partition low accuracy for traditional non-overlapping community detection algorithms,this paper proposes a non-overlapping community detection algorithm AMI-HC based on average mutual information.The algorithm performs community merging based on improved modularity increment,then determine the final community partition by calculating and selecting the maximum average mutual information value.Finally,compare the AMI-HC algorithm with other community detection algorithms(such as GN,FN,EO,LPA,CE algorithm)on real networks and artificial networks.The experimental results show that the AMI-HC algorithm has high accuracy of community partition.3)Aiming at the problem that poor stability and low accuracy of overlapping community detection algorithm COPRA which based on label propagation,this paper proposes an overlapping community detection algorithm AMI-COPRA based on average mutual information.The algorithm will guide the label selection of nodes based on the idea of maximizing the average mutual information value in the label propagation phase.Finally,the AMI-COPRA algorithm will be compared with other community detection algorithms(such as LFM,CFinder,SLPA,HMLPAi,COPRA,LPPB algorithm)on real networks and artificial networks.The experimental results show that the AMI-COPRA algorithm has highly accuracy and stability.Overall,experimental studies have shown that AMI methods are effective in non-overlapping and overlapping community structure.