Object-Level Data Model for Keyword Search over Relational Databases

Abstract

Keyword Search Over Relational Databases(KSORD) has been widely studied in recent years. However, existing KSORD methods are usually based on schema graph or data graph and they are actually tuple-level methods. That is, the retrieved objects are direct tuple-level relational data, and the retrieval results are tuple-connected trees which are difficult to be understood by end-users. There are still much work to do to further improve the effectiveness and efficiency of existing KSORD methods. The essential cause is that an entity is usually divided into some parts stored in different tables due to normalized relational database design. In fact, the relational data model is storage-oriented rather than end-user-oriented. Therefore, a novel method called Object-level Keyword Search Over Relational Databases(OKSORD) is proposed in this paper. In OKSORD method, relational data are modeled as an object-level data graph, in which each node may consist of several tuples to present the complete information of an entity. There are two key issues in OKSORD method, one is object-level data modeling for relational databases, the other is object-level searching and ranking based on object-level data graph. This paper mainly addresses the first issues. The main contributions are as follows. Firstly, the concept of OKSORD is introduced for the first time. Secondly, an algorithm for classifying relation schemas is proposed to partition relations into four categories: primary relations, secondary relations, linked relations and coding relations. Finally, an object-level data model for relational data is defined and the algorithm for generating corresponding object-level data graph is proposed.