关键词: 奇异值分解, 电子商务, 推荐系统, 过拟合, 协同过滤, 集成学习

Abstract: The collaborative filtering algorithm was improved to prevent overfitting in training rating prediction model. Experiments were put forth for  overfitting phenomenon  of the series of single value decomposition algorithms in rating prediction, and a method of adjusting parameters and iteration count to avoid overfitting phenomenon was proposed. The experimental results showed that this method could  find better rating prediction and avoid overfitting at the same time.

Key words: collaborative filtering;recommendation system, overfitting, ensemble learning, single value decomposition, electronic commerce