Social bookmark weighting for search and recommendation
由David Carmel, Haggai Roitman, Elad Yom-Tov所提出
發表於The VLDB Journal(2010)
--
Introduction
What is Social bookmarking?
給予社群中的document對應的註解tag,使得user可以有效利用整個社群中的資源,整個社群系統也能提昇其搜尋的效率和推薦的品質。
‧Tag recommendation:
Given a document, and the identity of the user who wishes to tag that document, recommend one or more tags to be used for annotating the document.
‧User recommendation:
Given a pair of tag and document, estimate who are the most likely users that would give the tag to the document, for purposes such as finding a community of people who are interested in a specific topic.
‧Document recommendation:
Given a pair of user and tag, estimate which are the most likely documents that would be given the tag by that user, for recommending new content to the user.
--
Notations
作者建立一個social bookmark的model,可以根據Tag, User, Document三種元素彼此的互動,計算出一個社群內bookmark的weight,接著分析此model帶來的效益。
系統可以由weight的高低來推薦user,或著提供user在搜尋上的幫助,同理類推,此weight也能反應在tag和document上面的高低。
(註:一篇document可以有多個不同tag和不同的user)
--
Evaluation
作者使用兩個不同的dataset來做實驗,並和related work所提到的常見方法做比較,然後使用MAP作為評估實驗好壞的方法。
Dataset:
‧Dogear - is an internal IBM enterprise social bookmarking system.
‧CiteULike - is an online bookmarking service that allows users to bookmark academic articles.
What is MAP(Mean Average Precsion)?
‧This measure evaluates a ranked list of entities by the average precision(AP)
--
Conclusion
實驗結果證明,新方法除了在CiteULike中做的tag recommadation實驗以外,其餘實驗和其他方法做比較的MAP都有提昇,作者給的解釋為,因為CiteULike為使用者較分散的社群,每位使用者之間的差異較大,針對同一類型的資料所標註的tag也較為不同,計算出來的weight就會較為分散,所以可能導致在tag recommadation上的提升效果較不明顯...
--
Feedback
如果把此方法應用在其他類似的領域,是否能得到有用的效果?
(旅行者←→旅遊訊息←→旅遊分類)
0 留言:
張貼留言