- 學位論文
實現多維度檢索暨標籤推薦之書籤分享系統
Enabling Multidimensional Retrieval and Tag Recommendation for Bookmark Sharing Systems
摘要
資訊超載的問題經常困擾人們，如何透過推薦系統協助使用者從大量物件中篩選感興趣的資訊，一直是一個重要的議題。本研究針對旅遊資訊提出一個旅遊書籤分享系統，其核心為標籤推薦的機制，藉由標籤的熱門程度、使用者對其他使用者的信任程度以及標籤在特定地區或月份的獨特性，分別產生幾組具特殊意涵的標籤列表，透過自動學習使用者點選列表的習慣，計算得到混合的推薦結果。在搜尋書籤時，系統提供多維度檢索介面，讓使用者選取多個標籤、地區與月份作為過濾條件。在效能評估上，我們邀請一群使用者在真實系統上進行操作，首先比較不同的推薦方法，分別觀察標籤數量及過濾條件數量對其反應時間的影響，並累計使用者對各方法推薦標籤的使用次數，結果顯示我們提出的信任標籤及特色標籤可達半數。在混合式的推薦上，自動學習式方法相較於其它兩種方法，各式各樣的準確率及排名量測均展現較佳的結果。
並列摘要
The problem of information overload often troubles people. How to help users select information of interest from a large amount of objects through recommendation systems has always been an important issue. This study proposed a travel bookmark sharing system for tourism information. Its kernel is a mechanism of tag recommendation. By the degree of tag popularity, the degree of trust from one user to another as well as the uniqueness of a tag in particular areas or months, several tag lists with special meanings are respectively created. Through automatically learning the habit of user clicking on the lists, a mixed set of recommendation results are computed. While searching bookmarks, our system provides a multi-dimensional retrieval interface, which allows users to choose multiple tags, areas and months as the predicates. In performance evaluation, we invited a group of users to operate on a real system. The different methods of tag recommendation were first compared by respectively observing the influence of the amount of tags and the number of predicates to the response times of these methods. In the meantime, we also accumulated the amount of tags used by the user for each method. The results showed that the trusted tags and unique tags we proposed were used by half. For the hybrid recommendation, the approach of automatic learning was compared with another two approaches and demonstrated better results in various measures of accuracy and ranking.