基于聚类融合的邮件社交网络社区划分方法  

Mail social network community partitioning method based on clustering ensemble

作　　者：张中军[1] 董仕[1,2,3] 

机构地区：[1]周口师范学院计算机科学与技术学院,河南周口466001 [2]华中科技大学计算机科学与技术学院,湖北武汉430074 [3]华盛顿大学(圣路易斯)计算机科学与工程系,美国密苏里州圣路易斯210023

出　　处：《云南大学学报（自然科学版）》2017年第2期178-184,共7页Journal of Yunnan University(Natural Sciences Edition)

基　　金：国家自然科学基金(U1504602;U1504603;61173170);河南省教育厅科技攻关研究(14B520014);河南省科技厅科技计划(162102310590;162102310147);教育教学改革项目(J2016037)

摘　　要：为解决邮件社区挖掘中涉及内容隐私及社区形态单一问题,提出了一种基于聚类融合的邮件社区划分算法.该方法中首先利用邮件社交网络特征及邮件属性衡量节点间距离,避免对邮件内容的分析导致涉及用户隐私,其次使用K-Means算法产生若干初始聚类结果,同时引入共协矩阵记录初始聚类时节点的归属,最后根据共协矩阵中邮箱节点间的相似程度,使用融合算法合并初始聚类结果得到最终社区结构.实验表明,该算法未使用邮件内容,得到的社区结构质量较高,并能发现多形态社区.In order to solve the problems related to user＇ s privacies and single forms in detecting mail communities ,the paper proposes an effective mail community partition method based on clustering ensemble.In a social network, the proposed method first utilizes the network structure features and mail attributes to measure the distances between nodes,instead of analyzing mail content that might be involved in user＇ s privacies.Then, the method generates initial clusters by the K-Means algorithm and adopts the co-association matrix to decide which cluster each node belongs to.Finally, according to the similarity between nodes in the co-association matrix ,initial clusters are effectively combined to produce final communities. Experimental results show that the communities obtained by this algorithm are of high quality, although it does not use E-mail content, also it can find communities with various forms.

关 键 词：聚类融合 社区划分 共协矩阵 最短路径 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]