模式无关的社交网络用户识别算法  

A Schema-Independent User Identification Algorithm in Social Networks

作　　者：叶娜[1,2] 赵银亮[1] 边根庆[2] 李健 何箐[1,2] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049 [2]西安建筑科技大学信息与控制工程学院,西安710055 [3]陕西电力信通有限公司,西安710075

出　　处：《西安交通大学学报》2013年第12期19-25,共7页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(61272458);陕西省自然科学基础研究计划资助项目(2013JM8021);西安建筑科技大学青年基金资助项目(2013JK1189)

摘　　要：针对识别社交网络用户时存在的模式不一致问题,提出了基于分块和二部图的用户识别算法.该算法通过将传统分块算法中的属性值精确匹配扩展为无模式信息下的属性值近似匹配,避免了传统用户识别时所需的模式对齐;使用加权二部图及Kuhn Munkres （KM）最大权匹配算法进行源用户档案与待匹配用户档案间的相似度计算,解决了用户档案间属性个数不同及语义语法异构的问题.在社交网站Profilactic上采集了965个用户的公开数据,采用召回率、精确率和综合指标等评价指标对算法进行了实验评估.实验结果表明,所提算法能够不依赖模式信息进行实例级跨系统用户识别,与基于属性值精确匹配的算法相比,所提算法的召回率提高了6.2％～9.5％,综合评价指标提高了3％～4.2％.A user identification algorithm based on blocking and bipartite graph is proposed to solve the problem of schema inconsistency in user identification across social networks. The schema alignment is avoided by extending the precise attributes matching in traditional blocking to approximately matching without using schema information. The weighted bipartite graph and the Kuhn Munkres （KM） algorithm are employed to calculate the similarity between the source and the candidate user profiles so that the problems of different attribute numbers as well as the semantic and syntactic heterogeneity between user profiles are solved. Public data of 965 users are collected from the Profilactic website and the proposed algorithm is evaluated using the recall, precision and F-measure metrics. Experimental results show that the proposed algorithm can implement cross-system and instance-level user identification without the use of schema information, and a comparison with the method using precise attributes matching shows that the recall of the algorithm is improved by 6.2 %-9.5 % and the F-measure is improved by 3%-4.2%.

关 键 词：用户识别 二部图 实例匹配 跨系统个性化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]