完整後設資料紀錄
DC 欄位語言
dc.contributor.author周運城en_US
dc.contributor.authorChou, Yun-Chengen_US
dc.contributor.author劉敦仁en_US
dc.contributor.authorLiu, Duen-Renen_US
dc.date.accessioned2014-12-12T01:53:24Z-
dc.date.available2014-12-12T01:53:24Z-
dc.date.issued2011en_US
dc.identifier.urihttp://140.113.39.130/cdrfb3/record/nctu/#GT079864512en_US
dc.identifier.urihttp://hdl.handle.net/11536/48621-
dc.description.abstract許多研究顯示推薦系統已經廣泛成功地運用於許多不同的領域中,它可以將使用者可能感興趣的項目,主動地推薦給使用者,讓使用者可以快速地在大量資料中得到自己真正需要的資訊,避免資訊過量,達到個人化資訊過濾的目的。 在眾多的推薦方法中,協同過濾方法利用其他使用者的意見來預測目標使用者對於物品的喜好,相當適合在社群網路以及電子商務網站的環境中使用。近來又有學者提出利用使用者之間的朋友關係以建構信任網路,並且結合協同過濾的推薦方法,這個構想除了提高推薦方法的準確率與覆蓋率之外,更可以確保推薦系統不被惡意的使用者攻擊,進而影響這個推薦系統的可信度。可是目前的研究中,對於如何有效整合顯性朋友關係信任以及隱性物品喜好信任的研究仍有不足,也沒有探討使用朋友信任關係在cold-start使用者與heavy-rater使用者情況下對於推薦準確率的變異與影響。 在這篇論文的研究中,我們提出一個新的信任網路協同過濾推薦方法,在我們的方法中,整合了顯性朋友關係信任與隱性物品喜好信任,進而導出一個使用者的信任程度來建構出信任網路,並且,我們採用一個動態的方法來調整這兩種信任度之間的相對重要性;除此之外,我們也提出如何擴展信任網路以便找出更多有效的鄰居。我們利用Epinions資料集進行實驗,結果證實我們提出的方法與傳統方法相較之下,可以在信任網路中找尋到更多有效的鄰居以進行推薦,並且可以更加有效的提升推薦方法的預測準確率以及覆蓋率。同時,在實驗中我們也證明,傳統只考慮朋友關係的信任網路方法雖然在cold-start使用者可以有好的推薦準確率,但是在heavy-rater使用者的情況下,還是需要考慮物品喜好信任度的推薦方法才能更有效提升推薦準確率。zh_TW
dc.description.abstractRecommender systems are successfully applied to many fields. Items that users may be interested in are recommended automatically. Therefore, users can quickly obtain personalized information from huge data and avoid information overloading. Among recommendation approaches, collaborative filtering (CF) predicts user interests of items merely based on user opinions, and is quite suit for social network and e-commerce services. Recently, some studies utilized trust network that are composed of friend relations to improve the accuracy and coverage of conventional CF. Additional benefit is preventing attack of malicious users, thus preserves the reliability of recommender system. However, there are no complete evaluation on the recommendation effectiveness of combination of explicit friend relation trust and implicit rating-based trust. Moreover, previous studies do not consider the variation of prediction accuracy with respect to different types of user such as cold-start user and heavy raters. We propose a novel trust-network CF recommendation approach, where trust network is constructed by the hybrid of explicit friend trust links and implicit rating-based trust links. Then, the trust network is extended by using rating-based trust links to discover more effective neighbors. In addition, we apply a dynamic method to adjust relative importance of explicit and implicit trust links. Experiments on Epinions dataset show that our approach outperforms conventional approaches in terms of the number of effective neighbors found and recommendation precision and recall. The results also show that conventional approaches have good performance for cold-start users, but fail to handle the existence of heavy raters properly.en_US
dc.language.isoen_USen_US
dc.subject推薦系統zh_TW
dc.subject協同過濾zh_TW
dc.subject朋友關係zh_TW
dc.subject評價信任zh_TW
dc.subject信任網路zh_TW
dc.subjectRecommender Systemen_US
dc.subjectCollaborative Filteringen_US
dc.subjectFriend Relationen_US
dc.subjectRating-based Trusten_US
dc.subjectTrust Networken_US
dc.title基於信任網路的協同過濾推薦方法之研究zh_TW
dc.titleA Study of Collaborative Filtering Recommendation Approaches based on Trust Networken_US
dc.typeThesisen_US
dc.contributor.department管理學院資訊管理學程zh_TW
顯示於類別:畢業論文