一种基于派系过滤的社区进化发现研究

Research on Community Evolution Discovery Based on Clique Percolation

在对派系过滤方法及其相关原理进行研究基础上,分析了该方法在社区进化发现中存在的参数依赖问题,提出了一种基于派系过滤的社区进化发现方法:通过生成社区树,综合多组参数的社区发现结果,可获取网络中不同耦合度的社区的层次结构,从而发现网络中社区的进化过程。本文将该方法应用在单词关联网络中,实验结果表明,该方法能够发现各社区在进化过程中的规模、成员以及耦合度方面的变化,在一定程度上,克服了传统派系过滤方法对参数的依赖性。

Clique percolation method has been used in many eases for discovering overlapping communities. However, its result is largely affected by parameters; as community size or cohesion changes the parameters needed shift accordingly. To overcome the dependency on parameters, in this paper we propose an approach for community evolution discovery based on community tree constructed by clique percolation. A community tree provides a hierarchical structure of communities discovered under a range of parameters in a given network; related community states can be found by searching a series of community trees thus the life span of a community can be discovered. We apply it to the word association network from DBLP data set and analyze how each community evolves. The outcome shows that this approach can effectively discover the community evolution process and identify the changes in size, membership and intensity. It＇s also observed that communities of different size have different evolving characteristic features.