基于可信度模型的重复主数据检测算法  

Duplicate Master Data Detection Algorithm Based on Credibility Model

作　　者：王继奎[1,2,3] 李少波[1,2] 

机构地区：[1]中国科学院成都计算机应用研究所,成都610041 [2]贵州大学省部共建现代制造技术教育部重点实验室,贵阳550003 [3]兰州商学院信息工程学院,兰州730020

出　　处：《计算机工程》2014年第5期31-35,40,共6页Computer Engineering

基　　金：国家科技支撑计划基金资助项目(2012BAF12B14)

摘　　要：针对来源于多个业务系统的重复主数据影响主数据质量、主数据同步及主数据挖掘等问题,提出重复主数据检测算法fastCdrDetection。从数据可信度的角度出发,在考虑数据源可信度、数据最后更新时间及数据长度的基础上,建立主数据可信度模型,并实现可信记录生成算法。设计非递归的字符串相似度计算算法FiledMatch,解决了由中文简写、缩写、错误拼写造成的主数据重复问题,采用sourceKeys算法对来源于同一业务系统、具有同样业务主键的重复记录进行预处理,从而提高重复主数据检测效率。通过对某电网基建物资63万余条供应商存量数据及23万余条模拟数据进行实验,结果表明,与PQS算法相比,fastCdrDetection算法的召回率由74%提高到88%,准确率由61%提高到95%,证明了算法的有效性。To avoid the effect of duplicate master data from multiple business systems on the quality, synchronization of the master data as well as master data mining, this paper propose a fastCdrDetection（Fast Cluster Duplicate Records Detection） algorithm, in which a duplicate master data detection model and a credible record generating algorithm are included, considering data source reliability, data refresh time and data length. A non-recursive algorithm FiledMatch is established for character string similarity calculation. Aiming at the eliminating problems caused by abbreviations and wrong spellings in Chinese input, a sourceKeys algorithm is constructed for pretreatment of duplicate records arising from a same business system and sharing same business keys to achieve high efficiency in duplicate master data detection. Experiments are carried on a power grid with 630 thousand records of raw material and 230 thousand simulated data records. Result shows that the recall rate of the fastCdrDetection algorithm is 88%, while the PQS algorithm is 74%, and the accuracy is 95%to 61%. The effectiveness of the algorithm is verified.

关 键 词：多数据源 重复主数据 可信度模型 检测算法 数据可信度 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]