Approximating Retrieval Unit for Un-structured Querying in Relational Databases

In recent years, there has been a significant interest in enabling keyword search capability, which is un-structured querying, for relational databases. Existing approaches return a set of tuples joined together as a retrieval unit. However, the joined tuples possibly involve data of the irrelevant or lessinteresting attributes for query keywords when the keywords are matched in multiple relations. This paper proposes a method to approximate dependencies as a heuristic to select the best set of attributes as the retrieval unit of keyword search in relational databases. We also discuss the advantages of our method in