Research Of Web Database Approximate Query Based On Semantic Similarity Computing

With the rapid expansion of the Internet, Web database has obtained the widespread application. More and more common users access the Web database by using the query interface to gain more information. Database query processing models have always assumed that the user knows what he wants and is able to formulate query that accurately express his query needs. However, most of the Web database users are common users. They often have insufficient knowledge about database contents and structure.As a result, their query intentions are usually vague or imprecise as well.And Therefore, the query results based on the user's query intentions often lead to empty or umsatisfactory answers. In order to reduce the user's tentative query times, or retrieve more satisfactory query results, the query submitted should act as approximate constraints for the query results. Therefore, the query should be an approximate one,in which the user's initial query can be extended through the use of related technologies. Also the approximate query can result in too many answers over large Web database, while the users are only interested in the results which meet their intention most closely. So it is important to rank the query results of the approximate queries.This thesis presents a novel approach of the approximate query and ranking (AQR) which depends on the original data from database and the query workload statistics to find the most similar values based on semantic similarity. The user's initial query criteria can be extended and the user's expectation information are then obtained as far as possible. AQR extends the categorical query criteria with the most similar values by evaluating the similarity of different pairs of values in the query workloads and expands the numerical query range to nearby values by using the kernel density estimation technology. AQR speculates the importance of each specified attribute based on the user's query and assigns the score of each attribute according to its "desirability" of the user, and then the relevant tuples are ranked according to their satisfaction degree to the user's needs and preferences.The thesis presents a approach to evaluate the user's preference to the unspecified attributes and solves the tuple ranking question of the same similarity. This approach further improves the ranking quality of the results of approximate query.The quality of the proposed approach is evaluated with experiments on two real databases. Here no domain knowledge or user's feedback is required in the whole process. The experimental results demonstrate that the proposed approach can capture the user's preferences effectively and have a high ranking quality as well.