关系数据库对象级别检索结果相关性排序算法研究

发布时间：2019-03-12 18:40

【摘要】：随着互联网的发展,网页搜索引擎获得了巨大的成功,用户可以用简单的关键词找到自己需要的信息。而关系数据库是当前数据库的主流形式,它采用结构化查询语言进行内容检索,并要求用户掌握一定的查询语言和数据库模式知识。由此就产生了一个很自然的需求,让关系数据库支持高效的关键词查询,因为关键词检索可以使用户摆脱SQL语句的束缚。相比于网络搜索引擎,关系数据库关键词检索又有了新的特点,例如：元组之间拥有语义关系；数据库中的属性值隐藏着等价和传递关系；数据库中的文本都是短文本等等。因此一些信息检索的方法仅是在关系数据库上进行元组级别的关键词检索,并不适合于关系数据库,需要研究一种适合于关系数据库本身特点的相关性排序算法。本文针对关系数据库的特点以及信息检索的特点,研究了一种对象级别的相关性排序算法。解决了元组级别检索排序的信息分散问题。本文的技术路线是：首先对关系数据库构建全文索引,按照模式图对数据库的元组进行信息整合,得到需要的对象；接下来在构建好的对象上进行关键词检索；最后针对检索出的结果进行相关性排序。本文提出的相关性排序算法首先需要发现属性值之间的传递关系。一个属性值出现的次数越多,属性值与关键词的联系越紧密,利用信息熵的方法为属性分配权值。信息熵的大小与数据分布的情况有关,可以通过计算信息熵来反映当前属性值分布的情况,找到属性值与关键词的相关情况,得到信息检索的相关性得分。其次需要考虑每个对象本身的结构特点。包括对象中的元组和元组之间的边的情况来得到数据库结构相关性得分,由两者共同得到相关性得分。本文采用上述方法设计了关系数据库对象级别检索结果相关性排序的总体框架,并实现了该算法。以手机领域的数据表为数据集对该算法进行验证,其结果证实了该算法的可用性以及算法的可行性。本文的排序过程不仅能得到包含关键词的对象信息,而且可以区分包含相同关键词的对象之间的差别；与传统的关键词检索排序算法相比,本文使用的方法能有效改善关系数据库关键词检索排序的效果。
[Abstract]:With the development of Internet, Web search engine has achieved great success, users can use simple keywords to find the information they need. Relational database is the mainstream form of database at present. It uses structured query language to retrieve content and requires users to master some knowledge of query language and database schema. As a result, there is a natural need for relational databases to support efficient keyword queries, because keyword retrieval enables users to get rid of the constraints of SQL statements. Compared with the web search engine, relational database keyword retrieval has new features, such as: there are semantic relationships between tuples; attribute values in the database hide equivalence and transmission relations; the text in the database is short text, and so on. Therefore, some information retrieval methods only do tuple-level keyword retrieval on relational databases, and are not suitable for relational databases. Therefore, we need to study a kind of correlation sorting algorithm which is suitable for the characteristics of relational databases. In this paper, according to the characteristics of relational database and information retrieval, an object-level correlation sorting algorithm is studied. The problem of information dispersion in tuple level retrieval and sorting is solved. The technical route of this paper is as follows: firstly, the full-text index of the relational database is constructed, the tuples of the database are integrated according to the schema diagram, and the required objects are obtained; secondly, the keyword retrieval is carried out on the constructed objects; Finally, the correlation order of the retrieved results is given. The correlation sorting algorithm proposed in this paper first needs to find the transitive relationship between attribute values. The more times an attribute value appears, the closer the relationship between the attribute value and the keyword is. The method of information entropy is used to assign the weight value to the attribute. The size of information entropy is related to the distribution of data. By calculating the information entropy, we can reflect the current distribution of attribute value, find the correlation between attribute value and keyword, and get the correlation score of information retrieval. Secondly, it is necessary to consider the structural characteristics of each object itself. The database structure correlation score is obtained by including the tuple and the edge between tuples in the object, and the correlation score is obtained by the two together. In this paper, we design an overall framework of relational ranking for object-level retrieval results in relational databases, and implement the algorithm. The proposed algorithm is verified by the data table in mobile phone field. The results show that the algorithm is feasible and available. The sorting process of this paper can not only get the object information containing keywords, but also distinguish the differences between objects that contain the same keywords. Compared with the traditional keyword retrieval sorting algorithm, the method used in this paper can effectively improve the sorting effect of keyword retrieval in relational database.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP311.13

【相似文献】