生物医学链接数据的清洗与集成技术研究

发布时间：2018-12-08 10:59

【摘要】：近年来语义网技术的高速发展方便了海量数据的集成与展示。生物医学领域由于其数据量大及子领域多的特性,清洗并集成各机构发布的RDF数据集的需求也日益凸显。过往的许多工作致力于应用语义网标准与技术为海量生物医学数据建立链接数据网络。例如采用语义网技术发布的生物医学数据集通常提供了指向其他数据集的交叉引用,但是这些引用往往存在错误,或是不能完整表达数据集间链接关系。已集成的数据需要通过使用SPARQL语言查询的方式来获取,阻碍了非语义网领域用户(例如生物医学领域专业技术人员)对数据的使用。各数据集使用不同的本体也使得跨数据集查询的结果难以集成。本文对生物医学数据集链接数据进行分析,并研究数据清洗及数据集成技术来解决上述问题。数据清洗技术对数据进行分析与校验,对重复数据,错误数据与缺失数据进行修正。语义网数据集成技术涉及本体匹配,实体链接等技术,其中本体匹配用于统一不同数据集本体的类与属性,实体链接连接不同数据集中指向同一实体的数据。本文的主要贡献如下:1.依托于Bio2RDF项目,调查并分析了主流生物医学链接数据。构建了数据集链接,实体链接及术语链接三类数据链接图,分析了链接图间关联性,发现了数据集链接具有小世界现象,实体链接度分布不严格符合幂次定律,不同数据集间术语有较多重合等现象。文章还通过研究实体链接属性,构建了一个标准测试集用于评估实体链接方法的优劣。链接分析方法可以通用于生物医学领域数据集分析;2.对选定数据集进行数据清洗,使用字符串检测,机器学习等方法对因为自动转换及人工输入产生的错误,补全缺失数据,修正错误数据,消除重复数据。同时根据实体链接的对称性和传递性分析补全数据集间缺失链接,修正错误链接,提升了数据质量及链接质量;3.在一个基于本体的数据集联合搜索引擎BioSearch系统中集成清洗后的数据集,使用本体匹配方法支持跨数据集联合查询。系统为用户提供简单高效的数据查询获取界面。实验结果表明使用联合查询及使用本文定义的语义查询接口比现有的两种链接数据搜索引擎更加高效,BioSearch所实现刻面过滤及实体浏览功能也被证实提升了用户体验。
[Abstract]:In recent years, the rapid development of semantic Web technology facilitates the integration and display of massive data. Due to the large amount of data and many sub-fields, the need of cleaning and integrating RDF data sets published by various organizations is increasingly prominent in the biomedical field. Many previous efforts have been devoted to the use of semantic Web standards and technologies to establish linked data networks for massive biomedical data. For example, biomedical data sets published using semantic Web technology usually provide cross-references to other data sets, but these references often contain errors or fail to fully express the link relationship between data sets. The integrated data needs to be obtained by using SPARQL language query, which hinders the use of data by non-semantic domain users (such as biomedical professionals). Different ontologies in different datasets also make it difficult to integrate the results of cross-dataset queries. This paper analyzes the linked data of biomedical data set, and studies data cleaning and data integration technology to solve the above problems. Data cleaning technology analyzes and verifies the data, and corrects the repeated data, error data and missing data. Semantic Web data integration technology involves ontology matching, entity linking and so on. Ontology matching is used to unify the classes and attributes of different datasets, and entity links connect different data sets to the same entity. The main contributions of this paper are as follows: 1. Based on the Bio2RDF project, the mainstream biomedical link data were investigated and analyzed. In this paper, three kinds of data link graphs, data set link, entity link and terminology link, are constructed, and the relationship between them is analyzed. It is found that the data set link has small world phenomenon, and the distribution of entity link degree is not strictly in accordance with power law. There is more overlap between different data sets. In addition, a standard test set is constructed to evaluate the merits and demerits of entity linking methods. Link analysis method can be used in biomedical domain data set analysis. 2. Data cleaning of selected data sets, string detection, machine learning and other methods to correct the missing data, correct the error data and eliminate the duplicate data caused by automatic conversion and manual input. At the same time, according to the symmetry and transitivity of the entity link, the missing link between the complete data sets is analyzed, and the error link is corrected to improve the data quality and link quality. 3. In an ontology-based data set federated search engine (BioSearch) system, the cleaned data set is integrated, and the ontology matching method is used to support cross-dataset joint query. The system provides users with a simple and efficient data query acquisition interface. The experimental results show that the joint query and semantic query interface defined in this paper are more efficient than the existing two linked data search engines. The facet filtering and entity browsing functions implemented by BioSearch have also been proved to improve the user experience.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13

【相似文献】