Research On Dynamic Network Community Detection Algorithm Based On Node Similarity

Community structure is an important topological feature of complex networks.Community detection has important theoretical and practical value for the research and application of complex networks.Because the node similarity calculation is simple and efficient,it has become the focus of many scholars’ research on community detection.However,in the existing community detection algorithm,the calculation of node similarity ignores the difference between common neighbor nodes,and needs to constantly adjust the parameters to obtain the optimal detection result.As nodes and edges are added and removed from the network,the community structure is constantly changing.Researchers have proposed corresponding dynamic community detection algorithms to capture the community structure in dynamic networks.Since the incremental-based dynamic community detection algorithm only considers the changing nodes and edges in the network,it greatly improves the efficiency of community detection in dynamic networks.However,the final result is affected by the initial network community structure and the incremental detecting process,which is prone to error accumulation.Therefore,aiming at the problems of community detection in static and dynamic network,this paper proposes a dynamic network community detection algorithm based on node similarity.The main research contents of this paper are as follows:For traditional static networks,in order to solve the problems of low differentiation of node pairs,complex selection of community aggregation parameters and randomness of community detection results,a community detection algorithm based on common neighbor clustering entropy node similarity is proposed.Firstly,in order to improve the community detection accuracy,the common neighbor and clustering coefficients are combined in the form of entropy to propose a new node similarity measure,which further accurately identifies the closeness similar node sets of each node.Secondly,in order to reduce the randomness of community detection results,based on the set of closeness similar nodes,the most closeness similar first-order neighbor nodes are determined by node leadership to merge and create initial communities.Finally,to address the problem of difficult parameter selection in the existing algorithms,a two-level merging approach is used to iteratively divide the final community by combining the idea of modularity optimization.The experimental results show that,compared with the other three algorithms,the normalized mutual information value of the proposed algorithm improves by 17.60% on average in static networks with real community structure,and the modularity value of the proposed algorithm improves by 7.56% on average in static networks that do not have real community structure.For dynamic networks,in order to solve the problem that existing incremental community detection algorithms ignore the changes of nodes and edges within the same community resulting in the accumulation of incremental detection errors,a dynamic community detection algorithm based on the similarity of edge-added nodes is proposed.First,a community detection algorithm based on the common neighbor clustering entropy node similarity is used to detect the community structure of the first snapshot network in dynamic networks.Afterwards,based on the community structure of the previous snapshot network,active nodes reflecting network changes are identified in the neighboring snapshots,and assigned communities for them to achieve the first stage of community detection.Then,in the incremental detection process,considering the impact of adding edges within the same community,the edge-added nodes were defined to split the community structure of the first stage into multiple local communities constructed by edge-added nodes and single communities.A series of community structures for the dynamic network are obtained by merging and optimizing the split communities.The experimental results show that the normalized mutual information value and modularity of the proposed algorithm improve by 43.63% and 74.34% on average in the synthetic dynamic network,and by 21.18% on average on the real dynamic network,relative to the other four dynamic community detection algorithms.