学习全局边函数的半监督社区检测  

Semi-supervised Community Detection Algorithm Based on Global Edge Function Learning

作　　者：丁彩英 李泽鹏 刘松华[1,3] DING Caiying;LI Zepeng;LIU Songhua(College of Information and Computer,Taiyuan University of Technology,Jinzhong 030600,China;School of Information Science and Engineering,Lanzhou University,Lanzhou 730000,China;School of Electronics Engineering and Computer Science,Peking University,Beijing 100871,China)

机构地区：[1]太原理工大学信息与计算机学院,山西晋中030600 [2]兰州大学信息科学与工程学院,兰州730000 [3]北京大学信息科学与技术学院,北京100871

出　　处：《太原理工大学学报》2019年第2期243-250,共8页Journal of Taiyuan University of Technology

基　　金：国家自然科学基金资助项目(61762047;61873178;61802158);国家重点基础研究发展计划(2013CB329600;2016YFB0800700);江西省教育厅自然科学基金项目(GJJ150686);江西省科技厅青年自然科学基金资助项目(20161BAB211015)

摘　　要：随着在线网络数据量激增,单纯分析网络拓扑结构、节点属性、边属性无法有效认识和理解其内在结构和特性,因此提出基于边函数的半监督社区检测算法。首先将拓扑结构和属性信息统一为先验知识,设计边函数便于引入属性等各类先验知识;在此基础上,结合传统半监督学习框架,采用半正定规划学习全局最优的节点归属矩阵。在人工合成数据、赣南客家数据和基准数据上的实验和分析表明,与已有传统半监督社区检测算法相比,该算法能有效利用各种先验知识,检测社区性能较好,并能较好地抵抗数据退化问题。With the explosion of the network data,it is difficult to effectively recognize the latent structure of the social networks by simply analyzing topology information,node attributes,and edge attributes.So,we proposed a semi-supervised community detection algorithm to give a unified respective of topology information and other attributes.First,the topology information and other attributes are taken as prior knowledge which is introduced and computed using the edge function.Then a semi-supervised framework is used to and learn the global node affiliation matrix by semi-definite programming.Experimental results on synthetic,Gannan Hakka data and benchmark network data show that our method can fully utilize all prior knowledge,the performance is better than that of many traditional algorithms and has better interpellation for detected communities.

关 键 词：社区检测 先验知识 赣南客家 半监督学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP393[自动化与计算机技术—计算机科学与技术]