| Summary: | 碩士 === 國立臺灣科技大學 === 電子工程系 === 90 === Since vast heterogeneous data provided on the Internet, web-visitors need search engines to find their required information. General-purpose search engines are able to return useful search results with respect to web-visitors’ queries. Therefore, how to rank web pages is critical to search engines. Although some search engines use their proprietary algorithms to rank web pages and try to increase the precision of search results, they still have the difficulty to meet web-visitors’ requirements. These difficulties include the lack of collaborative feedback and weak relationship between actual query and search results. Based on these reasons, in this thesis, we provide a new ranking method, named as HitRank, to include the user-oriented feedback in the ranking of search results. In this way, we can have higher opportunity to meet the web-visitors’ requirements.
This ranking method is based on constructing a directed graph referring to the navigation of click streams of web-visitors. An evaluative function is then applied to give a score to each search result clicked by web-visitors. It not only concerns with the times of a web page clicked by web-visitors, but also interests in the actual search behaviors of web-visitors. Based on HitRank approach, two types of search results, i.e. temporal and associated search results, will be given to web-visitors. The temporal search results present that different timeframes provide different search result orders. On the other hand, a web page with high HitRank score can provide an associated search results for other semantic-dependent search queries. Moreover, we develop a personal search portal based on proposed HitRank method. By this portal, web-visitors can find their required information quickly and improve the precision of search results.
|
|---|
