【作者】 李军；

【导师】 陈子阳； 管博；

【作者基本信息】 燕山大学 ， 软件工程（专业学位）， 2017， 硕士

【摘要】 实体匹配旨在找出多个数据源中表示同一实体的记录。实体匹配是数据有效使用的前提和关键,广泛应用于数据清洗、数据冗余检测、数据融合等领域。实体匹配的前提是了解记录之间匹配的规则,记录之间的匹配规则决定了实体匹配的质量,因此产生高质量记录匹配规则是实体匹配的关键。本文研究基于规则的实体匹配中记录匹配规则的产生算法。首先,通过对已有算法进行分析,发现已有算法不仅存在训练时间长的问题,而且存在结果精确率低的问题。针对存在的问题,本文提出一种基于贪心思想的记录匹配规则产生算法GR-Greedy。GR-Greedy先利用全属性产生基础记录匹配规则,再对基础记录匹配规则的属性进行排序,然后使用贪心的思想删除匹配规则中可能降低匹配精确度的属性,从而提高记录匹配规则的匹配精确度。其次,为了进一步提升记录匹配的精确程度,提出一种基于遍历思想的匹配规则产生算法GR-Traverse。与GR-Greedy相比,GR-Traverse在对基础记录匹配规则的属性进行排序的基础上,使用遍历思想扩大记录匹配规则的枚举空间,并删除匹配规则中所有可能降低匹配精确度的属性,避免了贪心思想的局部性问题,从而进一步提高记录匹配规则的匹配精确度。最后,通过真实数据集进行实验,从训练时间和匹配精确度等方面验证了本文算法的高效性和有效性。更多还原

【Abstract】 Entity matching is designed to identify the records of the same entity in multiple data sources.Entity matching is the premise and key of effective use of data,which is widely used in data cleaning,data redundancy detection,data fusion and other fields.Entity matching is based on the understanding of the matching rules between records,and the matching rules between records determine the quality of the entity matching.In this paper,we study the generation algorithm of record matching rules in rule based entity matching.First of all,through the analysis of the existing algorithms,it is found that the existing algorithms not only have the problem of long training time,but also have low accuracy.Aiming at the existing problems,this paper proposes an algorithm based on greedy thought to generate the record matching rules GR-Greedy.GR-Greedy uses the whole property to produce a basis of record matching rules,then on the basis of record attribute matching rules are sorted,and then use the greedy thought delete may reduce the attribute matching accuracy,so as to improve the accuracy of record matching rules matching.Secondly,in order to further improve the accuracy of record matching,this paper proposes an algorithm GR-Traverse based on traverse.Compared with GR-Greedy,GR-Traverse based on the basic record matching rules that attributes are sorted,using the traversal method to expand enumeration space,and delete all attributes that may reduce the matching accuracy,avoiding local problems of GR-Greedy,so as to further improve the accuracy of record matching rules matching.Finally,the effectiveness and efficiency of the proposed algorithm are verified by experiments on real data sets.更多还原