Improved Ranking Method for Keyword Queries on Relational Database

Abstract

Keyword search is an easy and potentially effective way to find information that is stored in relational database for ordinary users or web users. As results needed by user are assembled from joining tuples of multiple relations, ranking keyword queries are needed to retrieve relevant results by a given keyword query. For a given keyword query, we first generate a set of joining tuples, such as candidate networks (CNs). We then model the generated CN as a document. We evaluate the score for each document to estimate its relevance to a given keyword query. Finally, we rank the relevant queries by using each evaluated score as high as possible. In this paper, we propose a new ranking method by adapting existing IR scoring techniques based on the virtual document. We evaluate the proposed ranking method on DBLP dataset. The experimental results are shown by comparison of the proposed ranking method and the previous IR ranking method.