Research On Overlapping Community Mining Algorithm Based On Label Propagation And Multi-objective Optimization

A complex system is a complex whole composed of objects in the real world and the relations between objects.A complex network is an abstraction of a complex system.In a complex network structure,objects in the system represent nodes of the network,and the connections between objects represent edges.Network community structure is an important topological feature of complex networks,that is,the nodes within communities are closely connected,while the nodes between communities are sparsely connected.Overlapping Community Mining is used to discover overlapping communities in complex networks.With the development of the Internet,the overlapping rate of community structure in the complex network of human social is getting higher and higher,and the network community structure is becoming more and more fuzzy.How to excavate the accurate and stable community structure from the large complex network with fuzzy community structure has become a great challenge in complex network research.The accurate results of community mining can better reveal the potential structural characteristics and development trend of the real complex network from a mesoscopic perspective.The accurate results of community mining can better reveal the potential structural characteristics and development trend of the real complex network from a meso perspective,and then guide the problems existing in real life,which has a wide and important application significance in the neighborhood such as social contact,traffic,public opinion and even anti-terrorism.Due to the dynamic nature of the network in the real world,the algorithm of mining the overlapping communities in the static network cannot be directly applied to the dynamic network.Therefore,the algorithm of mining the overlapping communities in the dynamic network is also developed to solve the problem of mining the overlapping communities in the dynamic network.Label propagation algorithm is widely used in overlapping community mining because of its simple operation and approximate linear time complexity.However,as the topological structure of the real complex network becomes more and more complex,the community structure becomes more fuzzy.The existing algorithm of overlapping community mining based on label propagation shows strong randomness in the network with fuzzy community structure,which leads to low accuracy and multiplies the difficulty of overlapping community mining.In terms of community mining in dynamic networks,the current dynamic overlapping community mining algorithm based on decomposed multi-objective optimization has problems such as uneven distribution of community mining results in target space and sensitivity to proportional parameters,which leads to a decline in the performance of community mining.Based on the above two problems,two overlapping community mining algorithms are proposed in this paper,which are A label propagation algorithm combining eigenvector centrality and label entropy on the static network and Collaborative particle swarm with multi-objective optimization algorithm based on label propagation on the dynamic network.In order to solve the problem of high randomness and low precision of existing label propagation algorithms in the community structure fuzzy network,this paper proposes a label propagation algorithm combining eigenvector centrality and label entropy(ECLE-LPA).The algorithm first calculates the K-kernel iteration factor and eigenvector centrality of the nodes in the network,and then calculates the node influence.Through the node influence,a more accurate node ranking can be obtained to ensure the stability of the node order of label propagation.Then,based on the influence of nodes,the improved Rough Cores algorithm is used to initialize the label list of each node in the network and the corresponding node label membership degree.Then,based on the label entropy and the intimacy between nodes,the label list of the current node and the corresponding node label membership degree are updated to realize the label propagation,and then the redundant labels are deleted.At the end of each iteration,delete small communities surrounded by large communities,and continue iterating until the community structure becomes stable.Collaborative particle swarm with multi-objective optimization algorithm based on label propagation for dynamic overlapping community detection(CPSO-DOCD)is proposed to solve the problem of uneven distribution of overlapping community detection results in the objective space and the degradation of community detection performance due to the sensitivity of proportional parameters.In the algorithm,the reference point particle selection algorithm and collaborative particle swarm optimization algorithm are respectively used to solve these problems.The LPA algorithm is used to detection overlapping community.On the dynamic network structure,the algorithm uses a migration strategy to solve the problem that the results of community mining at the last time point are invalid in the current network structure.For ECLE-LPA algorithm,this algorithm is tested in real network and artificial network.In the real network,compared with the comparison algorithm,this algorithm improves the accuracy of the overlapping community mining in the EQ value.In artificial networks with fuzzy community structure,this algorithm has a better effect on NMI value than the comparison algorithm.For the multi-objective optimization algorithm based on collaborative particle swarm optimization that fused label propagation CPSO-DOCD,the experimental results show that: On the four dynamic networks of Cit-Hepph,Cit-Hepth,Emailed-eu-coretemporal and College Msg,the HV value of the overlapping community mining result set obtained by CPSO-DOCD was 0.2%-2% higher than that of the comparison algorithm.This indicates that the proposed algorithm is closer to the real Pareto frontier.On C-metric values,CPSO-DOCD performed better than the contrast algorithm.