一种基于函数依赖的属性相似度调整算法  

An Attribute Similarity Adjusting Algorithm Based on Functional Dependency

作　　者：谭明超[1] 刁兴春[1] 曹建军[1] 冯径[2] 

机构地区：[1]解放军理工大学指挥信息系统学院,南京210007 [2]解放军理工大学气象海洋学院,南京211101

出　　处：《上海交通大学学报》2015年第8期1075-1083,1089,共10页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金项目(61070714);解放军理工大学预研基金项目(20110604)资助

摘　　要：属性相似度的准确性是影响实体分辨准确程度的重要因素之一.为提高属性相似度的准确性,分析了属性相似度与函数依赖的关系,给出了属性相似度调整原则,提出了依据函数依赖进行相似度划分、相似度传递调整和计算相似度调整代价的方法,提出了通过属性相似度调整提高属性相似度准确性的属性相似度传递调整算法.实验结果表明,该算法能够更好地区分匹配记录对和不匹配记录对,获得更高的查全率、查准率和F1值.The accuracy of attribute similarity is one of the important factors affecting the precision of entity resolution（ER）.To improve the accuracy of attribute similarity,the relation between attribute similarity and functional dependency（FD）was analyzed and the principles for attribute similarity adjusting were suggested.The FD based methods for similarity partition,similarity transitively adjusting and cost computing of similarity adjusting were proposed.An algorithm for attribute similarity adjusting with FD（SAWFD）was put forward to improve the accuracy of attribute similarity.The experiment results show that the algorithm can better distinguish matching and unmatching records,and get higher scores of recall,precision and F1 measure.

关 键 词：实体分辨 属性相似度 函数依赖 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]