An Incremental SVD-based Scheme for Large-scale Recommendation Systems
Online shopping is becoming a common activity for more and more people in their daily lives. In an online store, numerous types of items are provided for users that results in large amounts of transaction and rating data. Recommendation systems analyze these past transaction patterns or rating data to provide personalized recommendations for users. Among several alternatives, matrix factorization algorithms such as singular value decomposition (SVD) can potentially be used for generating predictions in a more precise way. However, conventional SVD-based approaches suffer one serious limitation that recomputing the SVD of rating matrix is computationally very expensive. To adapt to the huge amount and rapid change of both user and item information in an online e-commerce environment, we thus aim at developing an incremental SVD scheme for generating up-to-date user-item rating predictions when streams of new rating data come in. Through our incremental scheme, the user-item rating matrix is directly updated by making use the existing result of SVD. Empirical studies show that our approach is efficient to be utilized in practical applications.
Incremental update large-scale data recommendation system
Kun-Fa Lin Wei-Guang Teng
Department of Engineering Science National Cheng Kung University Tainan, Taiwan
国际会议
2010 International Conference on Circuit and Signal Processing(2010年电路与信号处理国际会议 ICCSP 2010)
上海
英文
61-64
2010-12-25（万方平台首次上网日期，不代表论文的发表时间）