The Research Of Keyword Queries Over Relational Databases Based On Team

Relational databases are used to store structured data, a user must learn how to use SQL and have the knowledge of schema of the database, if he wants to query information over relational databases. It is a high threshold for the user. At the same time, keyword query is often used to retrieve information on the Internet, this is simple and easy to use, and is welcomed by common users. Thus, Keyword query over relational databases has attracted much interest recently.Most of works by now are aimed to obtain individual tuples to answer queries. We analyze these research works, and make an observation that use the combination of several tuples to answer queries will bring more value for users. So, this paper presents the concept of keyword queries over relational databases based on team. Here "team" means the combination of several tuples.We use inverted index to solve the problem. We think out a direct solution firstly:list all combinations of tuples in the inverted lists of query keywords, and then remove those teams which do not satisfy the restricted conditions in the user input, at last, remove the redundancy in the results. This Naive algorithm has a low efficiency, as it searches all of the solution space. Sometimes we can’t calculate the results because of high computation complexity.As the existence of the low efficiency in the Naive algorithm, we analyze the process deeply, make some important observations and bring forward several pruning strategies:remove the tuple which contains all of the query keywords, use restricted conditions earlier, and keep the optimal candidate. We reduce the amount of calculation as far as possible when we are integrating these strategies, and get a heuristic pruning algorithm. The pruning algorithm does not search all of the solution space any more, but search a part of it and add some judgment calculation, and thus reduce the computation complexity.We also add the mechanism of sorting the results. Some properties are observed and used to sort the results in a reasonable way, to make the results more acceptable to users. An empirical evaluation on both real data sets and synthetic data sets verifies that the pruning algorithm performances well in most cases.