Multi-Label Propagation Community Detection Algorithms Based On Node Influence And Information Entropy

Community structure is one of the important characteristics of complex networks.How to find community effectively is an important hot issue.Label propagation algorithm is a very important class of community discovery algorithms.Based on the classical label propagation algorithms LPA and MLPA,this paper uses the concepts of node influence and information entropy to improve and optimize the two algorithms and get better experimental results.In order to improve the randomness and irrationality of the update order of nodes in traditional label propagation algorithm LPA,this paper introduces the concept of node influence,combined with node similarity,proposes a new node influence calculation method,and uses it to improve the label propagation algorithm.In this paper,we propose three schemes that can be used to calculate the node influence(INF),and test these three different schemes on a small data set.We use the community number evaluation index(CN)to evaluate the results,and finally determine the optimal scheme.Based on this scheme,this paper proposes an improved INF-LPA algorithm.On the basis of this node influence calculation method,this paper also improves the multi-label propagation algorithm MLPA with the help of information theory.The improved algorithm can more accurately observe and discover the community structure in complex networks.Based on the multi-label propagation algorithm MLPA,this paper proposes a new community discovery algorithm AMI-MLPA by introducing the calculation of node influence and information entropy and taking the average mutual information(AMI)as a bridge.Firstly,an independent label is set for each node,and the propagation order is determined according to the node’s influence in the network.Then,in the process of label propagation,the algorithm combines the idea of average mutual information and the propagation intensity in MLPA to select the labels.By selecting the labels with higher average mutual information and propagation intensity,a more reasonable community division is obtained.Finally,the nodes labeled with the same label are divided into the same community.In this paper,experiments are carried out on real data sets and LFR benchmark synthetic data sets,and normalized mutual information(NMI)index is used as the evaluation standard.The results on real datasets show that AMI-MLPA algorithm is better than Fast GN,GN,LPA,MLPA,CDRS,DCN,GLPA,GLLPA.In particular,the accuracy of the partition results is 98.4% on a synthetic data-set with 100,000 nodes,which verifies the effectiveness of the algorithm.