Unlike relational database systems that return exact data items for a query, most current Web search and query systems return URLs of pages that might contain the required information. Processing the URLs to dig out the required information becomes a task of users, which is both ineffective and costly. In this work, we propose a novel Web query processing approach with learning capabilities. Under this approach, user queries are posted in the form of keywords that may not precisely describe the query requirements. A general-purpose search engine such as Yahoo! is employed to obtain candidate query results in the form of URLs. The first few URLs with high ranking are returned to a user for browsing. Meanwhile, the query processor learns how the user navigates through hyperlinks and locat...[ Read more ]

Unlike relational database systems that return exact data items for a query, most current Web search and query systems return URLs of pages that might contain the required information. Processing the URLs to dig out the required information becomes a task of users, which is both ineffective and costly. In this work, we propose a novel Web query processing approach with learning capabilities. Under this approach, user queries are posted in the form of keywords that may not precisely describe the query requirements. A general-purpose search engine such as Yahoo! is employed to obtain candidate query results in the form of URLs. The first few URLs with high ranking are returned to a user for browsing. Meanwhile, the query processor learns how the user navigates through hyperlinks and locates the query result within a Web page. With the learned knowledge, the query processor processes the rest URLs to produce precise query results without user involvement. The preliminary experimental results indicate that the approach can process a range of Web queries with satisfactory performance. The architecture of such a query processor, techniques of modeling HTML pages, and knowledge for navigation and queried segment identification are discussed. Experiments on the effectiveness of the approach, the required knowledge, and the training strategies are presented.