Research On Community Detection In Complex Network Based On Label Propagation

As one of the important characteristics of complex networks,community has attracted the attention of a large number of scholars at home and abroad.At the same time,community has also been applied to search engines,advertising recommendation,protein function prediction and other different aspects,which has great theoretical significance and application value.The purpose of community detection is to excavate the relationship between community(part)and the whole network(whole),predict the unknown functions of the network,discover the hidden rules,and grasp the future evolution trend of the network.At present,there are many community discovery algorithms,among which tag propagation algorithm is widely used in the field of community discovery because of its high speed and efficiency.Label propagation algorithm treats all nodes equally in the initialization stage,ignoring the difference of nodes.In the update stage,when the number of neighbor nodes is the same,it randomly chooses one.Because the algorithm does not consider the characteristics of nodes and adopts random selection strategy in these two stages,the stability and accuracy of the community detection algorithm for label propagation are poor.Based on this,this paper uses the importance of nodes in the network to differentiate the nodes,determine the initialization sequence,and reduce the randomness of the algorithm.Two different improved algorithms are proposed respectively.The main work of this paper is as follows:(1)The community detection algorithms are classified and summarized,and the representative algorithms in each category are listed,and the label propagation algorithm is introduced in detail.(2)A label propagation community detection algorithm based on community center(CN_LPA)is proposed.Firstly,from the local point of view,the concept of community center node is given by using the degree of node and the local centrality of aggregation coefficient.The community center node is used as the initial node to assign a unique label,and the neighbor node is given the same label as it.Then,the label is updated asynchronously and iterated repeatedly until the end of the algorithm.The experimental results show that the algorithm greatly reduces the randomness and instability of community partition results.(3)A label propagation community detection algorithm(NCEW_LPA)is proposed,which integrates the centrality and interaction of nodes.Since the CN_LPA algorithm only considers the local information of the nodes and ignores the influence of the nodes,NCEW_LPA algorithm is given in combination with the centrality and interaction of the nodes.Firstly,the algorithm calculates the centrality of all nodes based on the optimized LeaderRank method.According to the different interaction between nodes and their neighbors,the interaction force between nodes is obtained by similarity measure method.Then,the importance of all nodes is calculated by fusing centrality and interaction,and the sequence K with the largest importance of the first k is determined by means of mean and non-adjacent strategy.In label propagation update,the sequence K is centered on these nodes,and the importance is taken as the direction of propagation,and spread layer by layer until the end of the algorithm.Finally,the experimental results show that the algorithm can effectively identify important nodes and divide communities.