協同式過濾裡基於使用者匹配及物品匹配的轉移學習

Matching Users and Items for Transfer Learning in Collaborative Filtering

李重毅(Chung-Yi Li)

指導教授 ： 林守德

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.02963

摘要

本篇論文的研究目標是具有類似使用者和類似物品的同質性評分矩陣：即使不知道使用者的匹配關係和物品的匹配關係時，是否仍能在矩陣間進行轉移學習。更精確地說，我們假設有兩個評分矩陣，它們表達了同樣的喜好，而且兩個評分矩陣的使用者集合、物品集合都有很大一部份是重疊的。我們的目標便是找出這些使用者的匹配關係和物品的匹配關係，進而利用這樣的關係把一個矩陣的資訊傳遞到另一個矩陣上並改善其評分預測。
為了解出對應關係，我們會將稀疏的大型評分矩陣用低秩矩陣來近似，並用分解出來的因子來辨認使用者和物品。在解匹配問題的演算法中，一開始會把因子轉為奇異值分解的形式，並執行近鄰演算法。之後我們會指出奇異值分解的缺點，並用另一個目標函數來修正結果，以獲得更準確的匹配關係。最後，我們修改了協同式過濾常用的矩陣分解模型，使其能利用解出的匹配關係連結兩個矩陣，並做評分預測。我們的實驗顯示，在匹配問題中，我們能得到相當準的解。而即便匹配問題得到的解並不完美，我們仍能用其來改善評分預測模型。

關鍵字

協同式過濾 ； 轉移學習 ； 矩陣分解

並列摘要

This paper investigates the possibility of transferring information between homogeneous datasets of similar users and items but both user correspondence and item correspondence are unknown. More specifically, we assume there are two rating matrices that model the same kind of preferences, and there is a significant degree of overlap between the two user sets and between the two item sets. Our goal is to find out the user correspondence and item correspondence between the two rating matrices, and utilize the correspondence for exploiting the information of one matrix to improve the quality of rating prediction in the other matrix.
For finding out the correspondence, we factorize both rating matrices and exploit the latent factors to identify the users and items. The algorithm for solving the correspondence is initially based on singular value decomposition and nearest neighbor search, and then we point out the drawbacks of singular value decomposition and use another formulation to refine its result. Finally, we introduce a simple modification of regular matrix factorization model for transferring information across matrices with the obtained correspondence. In our experiment, we show that it is possible to solve the correspondence with decently high accuracy, and even with non-perfect correspondence obtained from our method, it is still possible to improve the quality of rating prediction.

並列關鍵字

Collaborative Filtering ； Transfer Learning ； Matrix Factorization

參考文獻

[2] Y. Koren and R. Bell. Advances in collaborative filtering. In Recommender Systems Handbook, pages 145–186. Springer, 2011.

[3] Y. Koren, R. Bell, and C. Volinsky. Matrix factorization techniques for recommender systems. Computer, 2009.

[5] B. Li, Q. Yang, and X. Xue. Transfer learning for collaborative filtering via a rating-matrix generative model. In Proc. ICML, 2009.

[6] P. Lops, M. de Gemmis, and G. Semeraro. Content-based recommender systems: State of the art and trends. In Recommender Systems Handbook, pages 73–105. Springer, 2011.

[7] H. Ma, H. Yang, M. R. Lyu, and I. King. Sorec: social recommendation using probabilistic matrix factorization. In Proc. CIKM, 2008.