資訊超載造成使用者的困擾,推薦系統遂成為幫助使用者篩選其感興趣資訊的工具。早期資訊推薦系統,因資料稀疏性及相似度計算耗時的問題,影響了推薦的準確性及實用性。近期推薦系統考慮使用者之間連結的關係,提昇了推薦的準確性,並朝個人化推薦發展。但對於大部分使用者,很難或很少有人可以快速且明確的決定與他人的連結關係,以獲得使用者真正感興趣的推薦資訊,對於無法決定關係的使用者,難以推薦使用者感興趣的資訊。有鑑於此,本論文分析使用者於操作物件行為上的關連性,找出其行為與使用者之間連結的關係,在使用者尚未明確確認彼此關係前,也可推薦使用者感興趣的物件,並結合社會性標籤產生多面向的信任關係,使得推薦物件符合使用者興趣。我們實做一推薦系統,基於社群網站上使用者連結關係,除了使分享資訊能夠快速地散播,更提昇個人化推薦的準確性。實際上線開放使用後,發現85%以上的推薦物件都能夠得到使用者認同。
Due to the problem of information overloading, recommendation systems have become useful tools for helping users to find the information they need. Most of the previous systems suffer from the data sparsity issue and the difficulty in similarity computation. Some recent systems take into account the trust relationships among users to improve the accuracy of recommendation. In this paper, we study the concept of trust and develop a framework of trust network, constructed by various kinds of user behaviors on common objects. Moreover, we integrate the social tagging mechanism into our framework and design a multifaceted trust-based recommendation method in which the trust between two users is distributed to individual tags. Based on this framework, we implement an online bookmark sharing system and make experiments on real users for evaluation. From the continuously accumulated results, we find that the users accept above 85% of the objects recommended by our method.