Community Detection And Evolution Analysis Of Dynamic Social Network

Dynamic community detection and community evolution analysis are challenging topics of social network analysis.For dynamic community detection,incremental clustering is one of the most efficient method,however,incrementally detecting community may result in partition error,and continuous error accumulation will lead to the deviation between the computed community structure and the ground-truth.For evolution events identification,core-node-based method is a simple and effective method,but this kind of method does not distinguish the contribution of core nodes to different evolution events,resulting in a reduced accuracy of evolution events identification.This paper propose corresponding methods to address these two problems.Firstly,we propose an error accumulation sensitive incremental community detection method for dynamic social networks.With the method,a snapshot will be totally re-partitioned once the error accumulation degree of incremental clustering exceeds a pre-defined threshold,rather than updating the community structure partially.Secondly,we propose an community evolution events identification method based on Superspreader and Superblocker.The method distinguishes core nodes into two kinds:Superspreader and Superblocker.Superspreader is closely related with influence maximization,and this kind of node effects the compactness of network components significantly.Therefore,Superspreader accounts for the birth,merge and growth of communities.Superblocker is intensively associated with network dismantling,and this kind of node effects the connectivity of network components significantly.Therefore,Superblocker accounts for the death,split and shrink of communities.Experiments on both real-world and synthetic networks show that not only the error accumulation sensitive incremental community detection method can efficiently detect the dynamic community,but also the community evolution events identification method based on Superspreader and Superblocker can accurately identify all kinds of evolution events,outperforming other methods from an overall perspective.