基于属性表的RDF数据存储系统研究

发布时间：2018-01-05 01:17

本文关键词：基于属性表的RDF数据存储系统研究　出处：《南京大学》2013年硕士论文　论文类型：学位论文

【摘要】：语义网是对当前万维网的扩展,通过在网页中加入语义内容,使得计算机能够更好地理解和处理这些信息。RDF是语义网中用来表示知识的数据模型,伴随语义网的发展RDF正得到越来越广泛的应用。随着RDF数据量的增加,如何高效地存储大量RDF数据成为一个难题。本文从RDF数据的属性表存储方法出发,研究针对RDF数据的属性选择算法、属性表动态调整算法及其相关技术。已有研究表明,不同的RDF数据集和查询往往需要不同的存储方案,现有的RDF存储方法在不同应用场景下各有优劣。属性表存储方法由于可以根据不同查询动态定制属性表结构,因此备受青睐。属性表存储需要指定一个属性选择算法,来确定各属性表中的属性,目前主流的属性选择算法大多直接套用其他领域类似问题的解决办法,例如数据挖掘中的关联规则挖掘算法、分布式数据库中的垂直分区算法等。本文提出一种新的针对RDF数据的属性选择算法,不仅可以根据查询动态选择属性表结构,达到为应用定制存储策略的效果,还针对RDF数据查询引入连接操作多的问题,优化了属性选择算法,尽量减少在表中的连接。在属性选择算法研究基础上,本文设计了可以在系统的运行过程中动态调整属性表结构的方法。由于属性表的创建和修改开销很大,现有的方法主要使用静态的属性表生成方法,不考虑属性表结构调整对系统性能造成的影响,这样的算法需要在系统启动前或者停机时运行。本文提出了一种判断系统负载的算法,该算法采用类似PID控制器的思想,可以根据各时间段内查询和响应的情况定性地给出系统是否空闲。并且本文设计了一种属性表的增量调整算法,可以在属性级别给出将旧表结构调整为新表结构需要进行的操作,并在每个属性调整前检查系统负载,只在空闲时进行调整,以将属性表动态调整的影响降到最低。最后,本文在现有的开源语义网工具Jena中增加了属性表存储的功能,通过修改其中的查询处理模块,可以将查询中合适的数据访问重定向到属性表上进行。为了考察在真实应用场景下属性表存储的性能,本文还在SP2Bench SPARQL性能测试工具中增加了模拟用户数变化的功能。结合改进的Jena与SP2Bench的实验表明,本文提出的属性表属性选择算法和调整时机选择算法可以显著地提高系统的性能。
[Abstract]:The semantic web is an extension of the current web, by adding semantic content in a web page, which the computer can understand and process the information.RDF is a semantic network knowledge representation model used in RDF, with the development of the semantic web is becoming more and more widely used. With the increasing amount of RDF data, how to efficiently the storage of a large number of RDF data has become a problem. This paper from the attribute table storage method of RDF data based on RDF data attribute selection algorithm, attribute table dynamic adjustment algorithm and its related technology.
Studies have shown that different RDF data sets and queries often require different storage schemes, existing RDF storage method have advantages and disadvantages in different application scenarios. The attribute table storage method because it can according to the different custom attribute dynamic query table structure, and therefore favored. Attribute table storage specifies an attribute selection algorithm to determine each attribute in the attribute table, the current mainstream attribute selection algorithm mostly directly applied to similar problems in other areas of the solution, such as association rules mining algorithm in distributed database vertical partitioning algorithm. This paper proposes a new attribute selection algorithm for RDF data, not only can query according to the dynamic selection the attribute table structure, reach for the application of custom storage strategy effect, but also for the RDF data query into the connection operation many problems, optimize the attribute selection algorithm Try to reduce the connection in the table as far as possible.
On the basis of property selection algorithm, this paper designs can be in the process of running the system to dynamically adjust the structure of the table. The attribute table creation and modification overhead, the existing methods mainly use the static attribute table generation method, without considering the attribute table structure adjustment impact on system performance, such as the algorithm needs before the system starts or stops running. This paper proposes a algorithm of load judgment system, the algorithm adopts PID controller similar ideas, according to each period of time, the query and response system is given to whether free. And this paper proposes an incremental algorithm for attribute table can be adjusted. The old table will be given at the attribute level adjustment for the new table structure need operation, and adjust the load before the check system in each attribute, only be adjusted when idle, to The impact of the dynamic adjustment of the attribute table is reduced to a minimum.
Finally, this paper adds the attribute table storage function in the existing open source semantic web tools in Jena, the query processing module to modify the query in data access can be appropriate to redirect the property tables. In order to investigate the performance in real application scenarios attribute table storage, based on the SP2Bench SPARQL performance testing tools. Increase of user number simulation function. The combination of Jena and SP2Bench show that the improved experiment, the proposed attribute table attribute selection algorithm and adjusting timing algorithm can significantly improve the performance of the system.

【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333;TP273

【相似文献】