异构数据联合式的真值发现算法  

Joint Truth Finding on Heterogeneous Data

作　　者：陈超 申德荣 寇月 于戈 

机构地区：[1]东北大学计算机科学与工程学院,辽宁沈阳110169 [2]渤海大学信息科学与技术学院,辽宁锦州121007

出　　处：《东北大学学报（自然科学版）》2017年第10期1373-1376,1387,共5页Journal of Northeastern University(Natural Science)

基　　金：国家重点基础研究发展计划项目(2012CB316201);国家自然科学基金资助项目(61033007;61472070)

摘　　要：互联网上提供的同一事实的信息通常会存在冲突,影响数据集成和知识发现.为了甄别真值,提出了一种基于距离的异构数据联合真值发现算法.首先,关于同一数据项,基于数据源声明值与真值的距离,计算数据项向量;采用KMeans聚类算法,获得数据项初始聚类.然后,迭代进行信任分析和聚类,即在每个类簇内,采用最优化思想,联合异构类型数据,更新事实的可信度和数据源的类簇内可靠性,重新计算每个数据项向量,再次聚类,迭代直至类簇达到稳定.实验结果表明:由于细粒度的数据源质量划分,联合考虑异构数据类型,可以获得更高的真值发现准确度.The value of an entity attribute on the web is usually provided by multiple data sources, but the values provided by them are not always the same,which affects the effective integration of data,so it is necessary to find out the true value among these given values. The existing truth finder algorithms mainly focus on the single type data kind, so a distance-based truth finding algorithm was proposed by considering heterogeneous data jointly. Firstly,for a specific data item, the data item vectors were calculated on the basis of the distance between the claimed value from every source and the truth value. The KMeans algorithm was used to get initial clustering. Then, alternate clustering and trust analysis were iteratively performed, i. e.,within each cluster, confidence of facts and trustworthiness of sources were updated with the idea of optimization and joint heterogeneous data. Each data item vector was recalculated and reclustered, and when each cluster was stable, the iteration would be terminated. The experiment results showed that the proposed algorithm has a higher accuracy for truth finding because of the fine grained partition of source quality and the joint model of heterogeneous data.

关 键 词：真值 真值发现 KMeans聚类 最优化 异构数据 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]