An SVD-based Collaborative Filtering approach to alleviate cold-start problems

Recommender systems, especially those based on collaborative filtering, help users filter large amount of unwanted information according to their own previous behaviors. However, most recommender systems encounter serious cold-start problems, which means such intelligent systems can hardly do anything to help users find out what they want when there is less or even no related information about user behaviors. This paper proposed an SVD-based Collaborative Filtering approach to alleviate such problems. One core idea behind this method is that lower-rank approximation could remove data noise brought by unstable user behaviors thus lead to better recommendation quality. Preliminary experiments show that the SVD-based CF approach not only improves the prediction accuracy but also has good performance.