一种从马尔可夫聚类簇发现潜在WEB社区特征的方法

Discovering Signature of Potential Web Communities from Clusters of MCL

在分析了目前一些典型的社区发现算法的基础上,通过对无主题条件下的隐含社区发现算法的研究,提出将基于流的社区特征和马尔可夫图形聚类算法(MCL)的簇结合起来寻找Web隐含社区的方法．将镜像或近似镜像页面的删除放在图形聚类之后,大大减少了比较的代价．然后,在聚类簇的基础上,使用判定每个簇内元素的筛选算法产生可能的社区候选集合．实验表明,该方法是可行的,可以发现许多存在的社区．

Web community is an important social activity in the evolution of Web. The paper analyzes typical algorithms of present Web communities＇ discovery. Under the condition of non-topic pre-defined and implicit communities, a new method is proposed, which combine both characteristic structure of community and the clusters of Markov Graph Clustering（MCL） to find implicit communities. The procedure of deleting mirror or near-mirror pages is arranged behind graph clustering so that decrease comparing cost considerably. Then a community member select algorithm is used to produce the set of community candidates. The experimental results show the new method works properly and many Web communities are inferred.