A Study On Performance Optimization Of Keyword-Query Over Relational Databases

In recent years, database technology has been widely used in various industry sectors, however, due to the rapid growth in the relational database users, structured query technique, which known as the classic query technique in relational database, has become an obstacle to non-professional users. For this problem, researchers have proposed a new query technology called Keyword Search Over Relational Databases(KSORD). This technique does not need user to master the structure of the query syntax does or to understand the structure and properties of the database, instead, with a few words from users, system can provide relevant information to users. just under will need to type a few keywords to query relational databases. Because of its ease of use, this technology has become a research hotspot.To users, current KSORD provide convenience, but researchers, there are a lot of challenges:First, information of a single result may not only from only one tuple, but in the form of debris scattered in a number of interrelated tuples which, according to a certain logic. The system is needed to transform these debris into conceivable information.Second, a query may produce a large number of results. How can the most user-interested results be found is also a problem.Third, what the user needs usually does not accurately reflected by the keywords in the query. How to help reconstruct query.The study also has some aspects of the work, but there are still problems on the efficiency and effectiveness. To address this issue, this paper altered existing datagraphs, designed a new Top-k result tree generation and sorting mechanism based on this new datagraoh, and developed relationship-based query term reconstruction method, All these techniques were applied to the prototype system called Extractor. Main contributions of this paper are the following: 1. A new undirected weighted datagraph is propose, with its structure adopting exist datagraphs and new defined node and edge weights.2. A set of function that used to execute query and generate results are developed according to the character of the new datagraph. These functions performed well in many aspects.3. The application to extend the semantics of keywords, such as "or" or "property+variable", is re-designed in this paper. In comparison to former ones, there is a obvious improvement in efficiency.4. A new Query-Reconstruction method is designed in this paper. Sometimes, query results are not satisfied. In this case, the user can choose to let the system reconstruct the original query and generate many new queries. Then the user will select a query that best reflect his need from these new queries, and through the implementation of new query to get different results. In this paper, two kinds of reconstruction is introduced, named query expansion and query correction. Query expansion:When the query is not specific enough to provide the results caused by the problem of low precision, the system can increase the number of query keywords and semantic complexity, and narrows results spectrum to improve precision; Query correction:When the query does not accurately reflect the user's intention, the system can modified the original query, changing the scope of the results to improve recall.