透過您的圖書館登入
IP:18.220.1.239
  • 學位論文

動態相似度協同過濾法的推薦系統

Dynamic Similarity Collaborative Filtering of Recommendation System

指導教授 : 李素瑛

摘要


在推薦系統的相關研究中,協同過濾法是其中一種最有效率的方法。然而,極少數的研究考慮到時間對於協同過濾法結果的影響。考慮時間因素的協同過濾法的研究中,多數是利用隨著時間衰減評分值來達到時間的目的。然而,因為評分值表示使用者的興趣程度,衰減評分值可能被誤解成人們對於物品的喜好,會隨著時間改變而有所改變,這與多數的事實不符。人們對於喜歡的東西,通常會維持著相同的看法,很少因為時間而改變。這樣的概念也可能造成推薦系統結果的誤差。 因此,我們提出了一個新的方法於動態協同過濾法上。與其衰減評分值,我們更確信衰減人與人之間的相似度是更為合理的想法。人們之間的關係,會因為時間改變了環境,改變人本身的興趣,而有所改變。大部分的人會與現在同處於相同工作或是念書環境的人們較為類似,而與舊朋友間的相似度,很有可能因為時間而有所改變。我們稱此方法為” 動態相似度協同過濾法”。此外,我們更提出了進階的應用,利用比較預測值與實際值的結果,能使每個使用者,在每個不同的時間點,都有著適合於個人的相似度衰退值。我們確信,每個人的相似度衰退是不會相同的,甚至同一個人在不同的時間點都會有所不同。因此,這樣的方法,不但解除了我們對於設定相似度衰退參數的設定問題外,更增加了預測的成功率。 在實驗的部分,我們提出了多種驗證的方法,證明我們的方法是更符合人們的行為,並且在執行時間上,有了很大的改進,使得我們的方法更適合實際上的用途。

並列摘要


In the researches of Recommendation System, Collaborative Filtering is one of the most effective approaches. With high accuracy in recommendations, however, few researches focus on Dynamic Collaborative Filtering which considers the time influence in Collaborative Filtering. This causes the recommendations inappropriate because the system might make a recommendation which is out of date. On the other hand, most of the existing dynamic Collaborative Filtering works are focused on Dynamic Weight. Dynamic Weight Collaborative Filtering uses decay ratings to achieve dynamic property. In other words, the rating might be multiplied by a decay weight according to the rating time. The older the rating is, the lower the rating becomes. Nevertheless, rating decay can also be interpreted as the changes of users’ favor. We believe that people would not actually change their perceptions on the same item because of time. Hence, we propose a different way in Dynamic Collaborative Filtering called Dynamic Similarity Collaborative Filtering (DSCF). The similarities among users are decayed rather than the ratings. In our opinion, we suppose that time might change the similarities among people. We also propose an enhanced method of DSCF. We feedback the predicted rating via actual value in order to obtain a more appropriate similarity decay rate. The experimental results demonstrate the proposed method has higher accuracy and less computation.

參考文獻


[1] G. Adomavicius and A. Tuzhilin, “Toward the Next Generation of Recommender System: A Survey of the State-of-the-Art and Possible Extensions,” IEEE Transactions Knowledge and Data Eng., vol.18, no.6, pp. 734-749, 2005.
[2] M. Balabanvic and Y. Shoham, “Fab: Content-based, Collaborative Recommendation,” Comm. ACM, vol.40, no. 3, pp. 66-72, 1997.
[5] R. Burke, ”Hybrid web recommender systems,” The Adaptive Web LNCS, vol. 4321, pp. 377–408, 2007.
[7] T. Chen, W. Han, H.Wang, Y. Zhou, B. Xu and B. Zang, “Content recommendation system based on private dynamic user profile,” Machine Learning and Cybernetics, vol. 4, pp. 2112 – 2118, 2007.
[9] Y. Ding, X. Li and M. Orlowska, “Recency-based collaborative filtering,” Proceedings of the 17th Australasian Database Conference, vol. 49, pp. 99-107, 2006.

延伸閱讀