Answer Aggregation for Keyword Search over Relational Databases

Abstract

with the amount of text data stored in relational databases growing rapidly, the need of the user to search such information is dramatically going up. Many existing approaches focus on finding a tuple matching a keyword query and return the result as a joining network of tuples of one or more tables. In this paper, we formulate an answer aggregation of keyword search over relational databases problem which merges related joining tuples from multiple tables to a single tuple to reduce redundancy in the results and improve the search quality. We developed an approach which exploits the tuple identity information for merging tuples rather than scanning throughout the results to find the common values. We further proposed a pruning algorithm which greatly reduces the number of redundant results after merging. We have conducted experiments extensively on two well-known databases (DBLP and IMDB). The experimental results show that the number of tuples in the results was dramatically reduced which noticeably improved the search quality while the merging time and the pruning time were relatively low when compared to the searching time.