采用链路聚类的动态网络社团发现算法  

A Community Detection Algorithm for Dynamic Networks Using Link Clustering

作　　者：董哲[1] 伊鹏[1] 

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450002

出　　处：《西安交通大学学报》2014年第8期73-79,共7页Journal of Xi'an Jiaotong University

基　　金：国家"973计划"资助项目(2012CB315901;2013CB329104);国家"863计划"资助项目(2011AA01A103;2011AA01A101);国家科技支撑计划资助项目(2011BAH19B01)

摘　　要：针对当前基于节点的动态网络社团结构发现算法难以发现稳定的社团结构的问题,提出了一种采用链路聚类的动态网络社团发现算法(LDC)。该算法首先从链路的角度得到网络的链路图结构;然后对比不同时刻的链路图结构,将动态网络中节点的添加与移除以及边的添加与移除等复杂的变化信息简化为链路添加和链路移除2种增量变化信息;再在前一时刻社团结构的基础上以改进的链路划分密度函数对增量变化信息中变化的链路进行处理,判断该链路是否加入到社团中从而得到最优的社团结构;最后将得到的链路社团转化成为最终的节点社团结构。实验结果表明,相比于当前基于节点的动态社团发现算法,LDC算法能够有效地发现网络中结构稳定的社团结构,其模块度值和标准化互信息值至少提高了0.19和0.13,且算法的运行效率要明显优于基于节点的动态社团发现算法。A community detection algorithm for dynamic networks is proposed to overcome the limitation that the current node-based dynamic community detection algorithm is difficult to identify the stable community structure.The algorithm uses link clustering technique and gets a link graph structure of the network,and then the complex incremental information in the dynamic network such as addition and removing of nodes and edges are simplified into addition and removing of links.An improved link partition density function is proposed to process a link in the incremental information and decide whether the link should be joined into the community based on the existing community structure to get the optimal community structure.At the end,the algorithm transforms the optimal link community structure into a node-based community structure.Experimental results and comparisons with the node-based community detection algorithm show that the algorithm can get the stable community structure and the modularity and NMI can raise at least 0.19 and 0.13,respectively,and that the efficiency of the algorithm is superior to the dynamic node-based community detection algorithms.

关 键 词：链路聚类 增量方法 社团发现 动态网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]