Research On Community Detection Methods Based On Different Kind Of Networks

Complex networks as an emerging science that can explain the existence of network phenomena and their complexity.Community discovery is an important element of complex network analysis,which helps people to understand the organizational structure and connectivity patterns of networks,and then guides practice.Therefore,community discovery research in complex networks is of great theoretical significance and practical value.Existing studies do not deal well with networks with obscure community structures,and treat the network edges occurring at different times as homogeneous edges,at the same time ignoring the phenomenon that the information carried by connected edges changes over time.It remains a challenge to further extend and refine the community discovery methods in different network types of data.According to the characteristics of static and dynamic networks,this thesis proposes a community discovery algorithm respectively.The specific research contents are as follows:Firstly,the thesis introduces the research background,the research significance,and the theoretical foundation of community discovery of complex networks.It reviews the research status of community discovery at home and abroad,attribute information in real networks,the model structure of complex networks,and the implementation method of Degree Corrected Stochastic Block Model(DCSBM).And it elaborates the research purpose and research contents.Secondly,the Community Discovery Algorithm of Complex Network Attention Model(CCAM)is proposed for networks with obscure community structure.The main innovations of the algorithm are as follows:(1)By introducing the attention model,it proposes a model that fuses three kinds of feature information of first-order,second-order neighbors,and modularity matrix,which can more sensitively extract the features of the network.(2)It proposes "complex network preprocessing-community game merging" community discovery combination,which can better compress the network and effectively reduce the running time of the algorithm.(3)Referring to the relative index of the unweighted network,it proposes a relative index applicable to the weighted network for balancing the community game process and optimizing the community division results.Thirdly,for the phenomenon that the information carried by connected edges changes with time,the thesis proposes the Bethe-based Dynamic Network Community Discovery Algorithm(BDCD).The main innovations of the algorithm are as follows:(1)In response to the phenomenon that the influence of events in the real world diminishes over time,the network event influence model is proposed,which integrates events at different moments to obtain the matrix of connected edge coefficients at the current moment.(2)By extending the Bethe-Hessian algorithm,we get the weighted Bethe-Hessian algorithm,which can be applied to the matrix of connected edge coefficients of dynamic networks for community division.(3)By introducing time attribute into the DCSBM model,we obtained the variable node dynamic random block model,and the generative network of the model can be used as experimental data for dynamic community discovery algorithms.The experimental results show that CCAM algorithm and BDCD algorithm perform well compared with other algorithms.In static networks,the CCAM algorithm can accurately and effectively classify communities by integrating information of three feature dimensions in the network through the attention model,and it outperforms traditional community discovery methods in networks with inconspicuous structures.In dynamic networks,the BDCD algorithm can maintain relatively excellent performance when the number of nodes changes significantly by synthesizing the influence information of events at different moments,and also has better experimental results in terms of normalized mutual information and modularity.