搜尋引擎幫助人們瀏覽網頁與尋找在全球資訊網相關網頁,然而搜尋結果排名往往是根據搜尋引擎廠商自己排名演算法所產生,與使用者真正需求常會有所出入,最有效改善搜尋品質的方法就是直接透過使用者回饋,而隱含回饋可以讓使用者不需改變搜尋行為,直接回應搜尋結果的好壞。 本論文提出一種新的相關性預測公式,利用長期性隱含回饋 (Implicit feedback)模式,從使用者的點選網頁順序可能性及搜尋引擎排名對使用者瀏覽行為影響的可能性,找出真正的相關性網頁,從蒐集使用者的瀏覽路徑 (User session)做為分析回饋資訊的依據,利用搜尋結果重排序 (Re-ranking)機制,重新回覆接近使用者需求的排名結果。本研究初步結果發現,隨著隱含回饋資訊增加,搜尋準確性確實能有效的提升。
Search engine helps people in surfing the web and finding relevant pages, however, the search results are ranked according to manufacturers’ specific search engine ranking algorithms, which is sometime opposite to the real demand of users. The most effective way to improve the search quality is to utilize the users’ feedback on the retrievals, and the emerging technique using implicit feedback allows the users to respond to the quality of search results without affecting the behavior of web surfing. In this thesis, we developed a new prediction algorithm for relevant pages. Using the long-term implicit feedback information, we are able to analyze the influence of user’s click order and the page rank provided by the search engine, so as to identify truly relevant pages. We collect user sessions for analyzing the feedback information, the retrieved pages are re-ranked as a response to the user’s demand. The preliminary experimental results discover that the precision rate of the retrieved result is increased as the information of implicit feedback is accumulated.