Ranking Keyword Search Results with Query Logs

Abstract

Keyword search provides a uniform way of accessing the vast volume of structured and unstructured data present in many enterprises. Existing research on improving the effectiveness of the keyword search has largely focused on result ranking mechanisms, with little consideration given to user feedback. We are working towards developing a new approach to ranking the results of keyword search over structured data (stored in databases) using feedback information in the form of query logs. Our work is based on the schema-graph-based approach to keyword search, which consists of a candidate network (CN) generation phase and a CN evaluation phase. Our proposal is to extract the frequent patterns from the query log of a user (or a user group) and use them in ranking the CNs generated by the first phase. We present a concise description of our approach and lay out our plan for the next stage of research. Preliminary experiment results on a real dataset are also included.