Research On Semantic Keyword Search Over Relational Databases

In recent years, keyword search over traditional relational database has been a hot topic in database region. For the keyword search over relational database, users can search information without knowing the database schema and utilizing complex SQL queries.With the development of the research, semantic keyword query becomes a research direction.This thesis systematically introduced research content and research status at home and abroad of keyword search over relational databases. And we briefly outline the main research directions, including data model, search algorithms, construction of the index and strategy of results sorting. Then we point out advantages and disadvantages of these measures, analyze the inadequacies of the existing research.A semantic keyword search strategy over relational databases is proposed in this thesis, which analyzes the semantics of the keywords not only using the schema of the database but also the dependencies among keywords. Thus our approach can also support searching text attributes and database metadata.This thesis puts forward a method of constructing weighted schema graph, which can better describe the semantic relevancy between relationships and improve query semantic accuracy. Then on basis of the weighted schema graph, an algorithm of candidate network generating which is based on breadth-first traversal is presented and a kind of candidate network evaluation model to sort the candidate network by their semantic relevance of the search is proposed. In the evaluation model, we use the relevance between the keyword and the relation which is estimated by Dirichlet smoothing technology as the node weight. The semantic distance is used to assign to the edge as its weight. Through the evaluation model, we can sort query results according to its semantics. This can provide users with a variety of semantics of the query results, and improve the experience of the users’search. This thesis presents a degree of attribute selection which mapping keyword in the attribute level.The experiments have shown the efficiency of semantic keyword search model we proposed, obtaining high recall and precision of keyword search and search performance.