Local Community Detection Based On Node Characteristics And Modularity Optimization

Complex network has community structure,which is a way to construct complex system in real data.For example,in social networks,the closely connected nodes often belong to a circle of friends,that is,a community.Therefore,local community detection in complex networks is the process of finding complete community where the seed point is located,and the seed point can be any node in the network.In recent years,more and more methods have been used to improve the accuracy of local community detection,but these methods still suffer from the following problems.They do not use the connectivity information in the network according to the local community characteristics,do not focus on the role of special nodes in the community,and do not deal well with low quality seed points with few edge connections,etc.To solve the above problems and improve the accuracy of local community detection,three local community detection methods are designed in this paper.The main contents are as follows:(1)A local community detection algorithm based on alternating cycle strong fusion and weak fusion strategy(ASFWF)is proposed.Among them,strong fusion can increase the modularity of local community,and in strong fusion,the algorithm proposes a new membership function.The weak fusion based on membership strength can integrate influential nodes into local community.To set the threshold of weak fusion correctly,the algorithm also proposes a local community evaluation index which does not need the real community division.The algorithm uses these two strategies to fuse nodes alternately.Specifically,it allows a weak fusion after each strong fusion.The use of alternating cycles allows both strategies to work at each stage of the algorithm,thus improving the accuracy of local community detection.(2)A local community detection algorithm based on improved higher-order modularity and edge information(HMEI)is proposed.Firstly,according to the motif degree of the seed point,the algorithm selects the first node to join the local community with different ways.Secondly,the algorithm improves the higher-order modularity function based on motif to extend the community and get the central part with tight connection.Finally,for the boundary of the community and the sparsely connected area,the membership strength between nodes and community is mined by edge information,to obtain more complete local community.The algorithm is effective for low-quality seed points,and experiments show that it is also effective for large-scale data sets.(3)A local community detection algorithm based on kernel node and seed point features(LCDKS)is proposed.Firstly,a new kernel node is defined as the node with the most abundant local higher-order information.And a kernel node expansion method is proposed.The kernel node expansion method runs through the whole process.Once the local community extends a kernel node,this expansion method is performed.Secondly,based on the characteristics of the seed point,this algorithm proposes the seed point expansion stage,which divides the seed point into three cases to expand and get the initial local community.Then the final local community detection result is obtained through the optimized extension stage and the community expansion stage.This algorithm focus on the characteristics of different nodes,makes it effective for different quality seed points,and uses kernel nodes to propagate community label to improve the accuracy of community detection.