基于距离的自适应Web数据库记录匹配方法  

Distance-Based Adaptive Record Matching for Web Databases

作　　者：崔晓军[1,2] 肖红宇[1] 丁立新[2] 

机构地区：[1]温州科技职业学院信息技术系,浙江温州325006 [2]武汉大学软件工程国家重点实验室,湖北武汉430072

出　　处：《武汉大学学报（理学版）》2012年第1期89-94,共6页Journal of Wuhan University:Natural Science Edition

基　　金：国家自然科学基金(60975050);高等学校博士学科点专项科研基金(20070486081);中央高校基本科研业务费专项资金(6081014)资助项目

摘　　要：Web数据库环境的重复记录识别是Deep Web信息集成的重要步骤,具有查询依赖性、缺乏训练样本、在线处理要求等特征,导致现有的实体识别技术无法适用.在分析现有方法基础上,引入动态属性权重调整思想,提出基于距离的自适应记录匹配算法,在计算记录对的相似度时,加大匹配记录集合中相似度较大的属性的权重,并加大非匹配记录集合中相似度较小的属性的权重,迭代处理从而达到自适应动态调整各个属性权重的目标.该方法不需要训练样本,也不需要人工参与,实验结果表明其适用于Web数据库环境的重复记录识别处理.One of the important steps of Deep Web information integration is identifying duplicate records over multiple Web databases.Due to the features such as query-dependency,the lack of training samples,and the online processing requirements,most state-of-the-art record matching methods are not applicable for the Web database scenario.Based on the analysis of the existing methods,an adaptive distance-based record matching method is proposed by introducing the idea of dynamic attributes＇ weights adjustment.In the iterative process of the calculation for the similarity of records,the weight of each attribute is dynamically recalculated by means of increasing the weights of the attributes with the bigger similarity in the matching records set and increasing the weights of the attributes with the smaller similarity in the non-matching records set.The proposed method does not require training data as well as human efforts and the experimental results show that it works well for the Web database scenario.

关 键 词：WEB数据库 记录匹配 实体识别 比较向量 权重向量 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]