| The keyword-based search has become the most important way that users acquire information from the Internet and a lot of relational databases constitute a major component of Deep Web, therefore researchers in the field are increasingly concerned about the research on keyword query over relational databases. With the development of the research, finding out the keyword query semantics as much as possible becomes a research direction.This paper proposes three basic semantics, including relation name semantic, attribute name semantic and value semantic, and two potential semantics, dependent semantic and sequential semantic are included. The Semantic keyword search model SKS is proposed in order to penetrate these semantics into query processes and is divided into offline module and online module.The offline module plays a role of providing supplementary for the online module by the information of databases and the query result logs. The structure of the keyword index and construction of the network graph are introduced in details. For different mapping for query keywords, the keyword index will return different structures, which reflects the basic semantic of keywords. The network graph is built relying on the schema graph and query logs and based on the Apriori algorithm, being used for generating candidate networks, by which the dependent semantic is reflected.The thesis explains candidate network generation, query conversion and result rank in online module. The algorithm of candidate network generating is based on single source optimal path, which finds networks by scanning network graph. It provides query templates to facilitate the query translation of candidate networks and ranks the query results by clustering and other operations and the process well explain the sequential semantic.The experiments have shown the efficiency of SKS we proposed, obtaining high recalls and precisions. At the same time, the correctness of basic semantics and potential semantics is demonstrated. |
PDF Full Text Request