Dynamic Overlapping Community Discovery Algorithm Based On Label Propagation

Community discovery is an important part of complex network research.The main purpose of community discovery is to mine a group of nodes with the same or similar characteristics in the network.The community has found widespread applications in real life,such as danger warnings,recommendation systems,and public opinion analysis.Most of the traditional community discovery algorithms are based on static community research and have accumulated a large number of experiments and conclusions.But the real world network is changing all the time.Similarly,the community structure in the network will evolve into different structures as the network changes.The community will generate,dissolve,synthesize,decompose,expand,and shrink.The members of the community will not simply belong to a certain community,they may belong to multiple communities at the same time.According to this analysis,the traditional algorithm about static network community discovery has been difficult to meet the current demand for community discovery in the complex network world.Therefore,research and exploration on the discovery of dynamic overlapping communities is imperative.First,this paper introduces the relevant theoretical techniques of community discovery,then explains the more representative algorithms for static and dynamic community discovery,and analyzes the advantages and disadvantages of each algorithm;Then analyzes an application called COPRA,which can be applied to overlapping community discovery in depth.In order to improve the instability and randomness of the COPRA algorithm in the process of label propagation,this paper introduces the concept of node entropy to measure the importance of nodes by introducing the concept of entropy,then proposes the concepts of label value,which based on three factors : node entropy,label membership coefficient and node degree.According to these,the paper proposes the CI-COPRA algorithm.The algorithm is experimentally verified on four natural data sets and two artificial data sets of different sizes,and then compared with four other classic community discovery algorithms.The results show that the algorithm has lower stability than the improved randomness.The improvement has a better effect on community discovery.Then,the proposed CI-COPRA algorithm divides the initial community,and the analysis is performed according to different types of network increments.The concepts of community core and community similarity are introduced to judge the community evolution,and a DCI-COPRA algorithm is proposed.Moreover,the paper proposes the concept of modularity change as a criterion for measuring the quality of network division by expanding the concept of global modularity.The network community at 9 moments was constructed on the synthetic data set and compared with other 4 dynamic community discovery algorithms.The results prove that the DCI-COPRA algorithm can accurately capture the community evolution of the network in various periods.Divided communities are of higher quality.