基于实体演化的记录链接算法

发布时间：2018-03-22 04:35

本文选题：实体演化　切入点：记录链接　出处：《南京大学学报(自然科学)》2017年06期 　论文类型：期刊论文

【摘要】：实体识别(Entity Resolution)是指判断一个或多个数据源中两个不同记录是否描述相同实体,它有时也被称作记录连接(Record Linkage),在数据集成中被用于数据清洗(Data Clean)、去重(Deduplication)和相似连接(Similarity Joins)等集成操作中.实体识别技术可被广泛应用于人口普查、引文识别、Web搜索、数据清洗以及剽窃检验等诸多领域.然而,在真实世界中,实体的属性会随着时间的变化而变化,两条记录的属性值不同不能表明这两条记录对应不同的实体,具有相同的属性值的两条记录也不能表明对应相同的实体.时间记录链接就是匹配描述同一实体的带有时间戳的记录.已有的解决时间记录链接的方法依赖于时间模型来捕捉实体的演化,但是已有的时间模型在预测实体的演化时,实体匹配准确率不高,而聚类计算复杂度较高.为此提出了更加细致的捕捉实体演化的模型和新的两阶段的快速聚类算法.通过在三个真实数据集上的实验结果表明,提出的时间模型可以更加细致地捕捉实体的演化,提出的聚类算法能更快速而准确的聚类描述同一实体的记录,提高了识别的准确率和效率.
[Abstract]:Entity identity Resolution means to determine whether two different records in one or more data sources describe the same entity, It is also sometimes referred to as record link link, which is used in data integration, such as data cleaning, data cleaning, and similar connection Similarity Joins. entity recognition technology can be widely used in census, citation identification and Web search. Data cleansing and plagiarism testing are a number of areas. However, in the real world, the attributes of the entities change over time, and the difference in attribute values between the two records does not indicate that the two records correspond to different entities. Nor can two records with the same attribute value indicate the same entity. A time record link is a record with a timestamp that describes the same entity. Existing methods for resolving time record links depend on time. To capture the evolution of entities, However, when the existing time models predict the evolution of entities, the accuracy of entity matching is not high. For this reason, a more detailed model to capture entity evolution and a new two-stage fast clustering algorithm are proposed. The experimental results on three real data sets show that, The proposed time model can capture the evolution of entities in more detail, and the proposed clustering algorithm can more quickly and accurately describe the records of the same entity, and improve the accuracy and efficiency of recognition.
【作者单位】：东北大学计算机科学与工程学院;
【基金】：国家自然科学基金(61472070,61672142)
【分类号】：TP311.13

【相似文献】