Research On Dynamic Overlapping Community Detection Algorithm On Large-Scale Heterogeneous Information Network

Networks in the real world tend to change dynamically over time,which can lead to changes in the network community structure.Real-time community detection on dynamically changing networks to acquire the evolution process of communities on the network is of great significance for analyzing the impact of entities changes and changes in the relationships between entities in the network on the network structure.The existing dynamic network community detection algorithms are mostly time-complex and cannot be applied to community detection for large-scale dynamic networks.The incremental based dynamic network community detection method has low time complexity,but the community detection result is not accurate because the method of processing the increment is not perfect.In addition,the existing incremental-based algorithm has the problem of consistency,that is,since only the local community structure change of the network is considered when processing the increment,the global community structure of the network cannot be guaranteed to be reasonable over time.Finally,the existing methods cannot detect the community structure on the heterogeneous network,while in reality the networks are mostly heterogeneous.In order to detect the community structure on heterogeneous dynamic networks quickly and accurately,in this paper,a dynamic community detection algorithm based on incremental analysis and multiplex network extraction for heterogeneous information network(HIAME)is proposed.Firstly,we propose an improved incremental-based dynamic homogeneous information network community detection algorithm HomoIA,based on the comprehensive consideration of all network incremental types,we redefine the membership indicator for nodes and improve existing methods to improve the accuracy of the incremental-based homogeneous community detection algorithm.Then by introducing the concept of multi-path network,HomoIA is extended to the heterogeneous information networks.Finally,we solve the problem of consistency by defining global monitoring.In this paper we uses Microsoft academic network,Yelp network and LFR benchmark network as the experimental data set to verify the effect of the algorithm.We uses the modularity Q,the normalized mutual information NMI and the number of detected communities to evaluate the performance of HomoIA and HIAME.We use graphical tools to demonstrate community evolution behavior in the network detected by the algorithm.The experimental results show that the algorithm can improve the quality of community structure detected by existing incremental-based algorithm and can also detect various possible community evolutions in homogeneous and heterogeneous information networks.In this paper,we first introduce the research background and research status of dynamic information network community detection,we summarize and analyze the problems existing in the current research.Then we elaborate the dynamic homogeneous information network community detection algorithm HomoIA and the dynamic heterogeneous information network community detection algorithm HIAME.Finally,we demonstrate and evaluate the effects of the proposed algorithms through experiments.