从链接密度遍历序列中挖掘网络社团的层次结构  

Mining Hierarchical Community Structure Within Networks from Density-Connected Traveling Orders

作　　者：黄健斌[1] 孙鹤立[2] Dustin BORTNER 刘亚光[1] 

机构地区：[1]西安电子科技大学软件学院,陕西西安710071 [2]西安交通大学计算机科学与技术系,陕西西安710049 [3]Department of Computer Science,University of Illinois at Urbana-Champaign,Urbana 61801,USA

出　　处：《软件学报》2011年第5期951-961,共11页Journal of Software

基　　金：国家自然科学基金(60933009);陕西省自然科学基础研究计划(SJ08-ZT14)

摘　　要：提出一种称为TRAVEL的网络聚类算法.它能够产生包含所有可能密度聚类的网络链接遍历序列,并从中自动发现网络的全局优化聚类.然后,遍历序列被转换为连续子区间堆结构.在此基础上,提出一种聚类算法HCLU,可以无须用户干预地从连续子区间堆中自动发现网络的层次聚类边界.在真实网络以及计算机生成的仿真网络数据集上的实验结果表明,所提出的算法比目前的基准方法具有更高的聚类精度.此外,算法能够从各种带有噪声的网络中发现无冗余且鲁棒的层次社团结构.This paper proposes a density-based network clustering algorithm, TRAVEL. The algorithm produces a traveling order containing clustering with various densities and finds the optimal clusters in it. The traveling order is subsequently transformed into a data structure of contiguous subinterval heap based on which a clustering algorithm, HCLU, is designed to find the hierarchical cluster boundaries of the network without any user interaction. Experimental results on real-world and computer-generated synthetic networks show that the clustering accuracy of the proposed algorithms is higher than the baseline methods. Furthermore, they are able to produce robust hierarchical communities in various networks with low redundancy in the presence of noise.

关 键 词：基于密度的网络聚类 层次社团发现 中心点 离群点 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]