Research On Object-Level Retrieval Algorithm Over Relational Databases

Massive structured data is stored in the relational databases, and the amount of data is still growing every day. With the maturity of database technology and information retrieval technology, it makes the user can retrieval structured data easily and efficiently in the database, keyword query over the database is a hot research topic recently. This technology allows the users needn’t understand the complex structured query language and the underlying schema of the databases, they will be able to query data in the databases. The current research work is mainly based on the tuple-level methods. However normalized relational databases design result in that the complete information of entities are usually divided into some parts stored in different tables, as a result, the results returned by tuple-level methods are not complete information, and are difficult to be understood by ordinary users.Object-level information retrieval methods are more intuitive to describe the semantics of the search results, and the returned search results have more context information, and they are more in line with the needs of the user’s query. Object-level information retrieval methods have attracted more and more attentions and research focuses. The main contributions in this paper are as follows. Firstly, the present several typical retrieval methods with object-level thinkings and perpectives are deeply analyzed. Secondly, a preliminary definition of the object model for relational databases is proposed. In brief, an object is the set of closely related tuples in the database. Thirdly, an object-level information retrieval method over relational database is designed, this method is called DBORank. A flexible and effective scoring mechanism is designed from both databases and information retrieval perspectives, which takes into account not only the link structure of object-level data graph, but also the internal structure of object nodes, the type and weight of edges, the relevance of object content, and et al. DBORank method improves the retrieval effectiveness. At the same time, the iterative algorithm and the sorting algorithm optimizes the retrieval efficiency of DBORank method.Finally, an object-level information retrieval prototype system over relational databases is implemented. We use the DBLP data set to verify the retrieval effectiveness and efficiency of DBORank method by experiments. In this paper two evaluations are employed to evaluate the retrieval effectiveness of DBORank method, they are P@K and Mean Average Precision respectively. Mean while the retrieval efficiency of iterative algorithm and Top-K sort algorithm of improved method were tested. Extensive experiments show that DBORank method performs effectiveness and efficiency.