面向中文知识图谱的数据重组与应用

发布时间：2018-09-10 18:30

【摘要】：随着语义网的快速发展,越来越多基于RDF的图谱数据被发布到万维网上,组成了开放链接数据(Linking Open Data)。一般来说,这些开放数据提供SPARQL查询服务和关键词访问服务。实际上,相当一部分用户在访问的时候会选择关键词访问,这些访问行为同时也被记录在服务器的日志中。尽管用户期望进行表达能力更强的查询,SPARQL的复杂性和对所要查询图谱的不了解,往往会使得用户很难获得理想的查询结果。除了RDF这种语义网数据交换的标准格式之外,随着No SQL的兴起和发展,基于属性图数据的查询和存储也得到越来越多的关注和研究。虽然有一部分基于属性图的评测标准已经发布并且应用到实际的场景中,但仍然缺乏被广泛认可的衡量综合性能的评测基准。因此,如何更好地组织和使用迄今为止积累下来的海量基于RDF的语义数据,已经成为语义网领域一个开放性的问题。毕业设计正是从这个背景出发,提出了一个针对RDF的图谱上SPARQL查询推荐的框架和一个利用RDF数据对属性图进行基准评测的方法。具体来说,首先是提出来一个针对SPARQL查询进行推荐的框架。该框架是通过分析知识图谱的访问日志,挖掘得到用户查询的偏好情况,并结合用户的原始SPARQL查询语句,推荐合适查询语句。Zhishi.me上的实验结果表明推荐后的查询语句能返回具有更好可读性的查询结果,能帮助用户更好地使用SPARQL语句来遍历知识图谱。除此之外,本文还提出了一个利用已有RDF数据集来生成评测属性图的存储的评测基准。该方法先将RDF的数据模型转换为属性图的数据模型,并通过分析访问日志来生成相应查询语句集。基于Zhishi.me的数据集实现了该评测基准,并对Neo4j和Titan这两个目前最流行的支持属性图存储的数据库进行了充分的评测,为用户选择使用合适的数据库提供了可靠的参考依据。
[Abstract]:With the rapid development of semantic web, more and more map data based on RDF have been published on the world wide web, forming open link data (Linking Open Data). In general, this open data provides SPARQL query services and keyword access services. In fact, quite a number of users choose keyword access when accessing, and these access behaviors are also recorded in the server log. Although the complexity of Sparql and the lack of understanding of the query graph, it is difficult for users to obtain ideal query results. In addition to RDF, the standard format of semantic web data exchange, with the rise and development of No SQL, query and storage based on attribute graph data have been paid more and more attention and research. Although some attribute graph-based metrics have been published and applied to actual scenarios, there is still a lack of widely accepted benchmarks to measure comprehensive performance. Therefore, how to better organize and use the accumulated mass of semantic data based on RDF has become an open problem in the field of semantic Web. It is against this background that the graduation project proposes a framework for SPARQL query and recommendation on the RDF graph and a method for benchmarking attribute diagrams using RDF data. Specifically, the first is to propose a framework for the recommendation of SPARQL queries. By analyzing the access log of the knowledge map, the framework can mine the preference of the user query, and combine with the original SPARQL query statement of the user. The experimental results on the recommended query statement. Zhishi.me show that the recommended query statement can return the query results with better readability and can help users to traverse the knowledge map better by using SPARQL sentences. In addition, this paper proposes a benchmark to generate the store of the attribute graph by using the existing RDF data set. Firstly, the data model of RDF is transformed into the data model of attribute graph, and the corresponding query statement set is generated by analyzing the access log. The data set based on Zhishi.me implements the benchmark, and gives a full evaluation of the two most popular databases, Neo4j and Titan, which support the storage of attribute diagrams, which provides a reliable reference for users to choose the appropriate database.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3

【相似文献】