摘要： 社团检测简化是重要的图挖掘问题,动态网络上的重叠社团检测及其社团演变模式是近年来的研究热点,但将静态网络的局部优先的社团检测算法应用到动态网络的重叠社团及其演变模式检测上的研究较少。为此,提出一种局部优先的动态网络重叠社团演化分析方法。该方法在每个网络快照上利用标签传播算法检测局部Ego社团,通过不断合并局部Ego社团得到全局社团结构。在此过程中引入社团相似度与关联度2个概念,利用演化聚类框架进行重叠社团演化分析。实验结果表明,该方法不仅能有效地发现动态网络中重叠社团结构,而且还可以分析出社团随时间的演化模式。

关键词: 动态网络, 重叠社团, 局部优先, 演变模式, 时序平滑框架, Ego社团

Abstract: Community detection is an important image mining problem,and the overlapping community detection and the community evolution pattern in dynamic networks is a hot topic.However,There is no work on the integration of the local-first community detection.This paper proposes a local-first analytical approach to analyze the evolution pattern of the overlapping communities on dynamic networks.This approach detects the local Ego community via label propagation on each network snapshot.Then,the global communities are generated via merging the local Ego communities successively.In this process,the two concepts of the community similarity and community association are introduced,and it analyzes the overlapping community evolution pattern via an evolutionary clustering framework.Experimental results show that this approach not only detects the overlapping communities in dynamic networks effectively,but also analyzes the evolutionary pattern.

Key words: dynamic networks, overlapping community, local-first, evolution pattern, timing smoothing framework, Ego community

中图分类号: 

• TP18