In this paper,a fusion of singular value decomposition and clustering algorithm named SVD biKmeans collaborative filtering (SBKCF) is proposed, in order to solve the scalability and sparsity problems in user based collaborative filtering system. The algorithm adopts improved Pearson similarity metric formula to measure similarity between users, and clustering those users which have been dimension reducted, then generates the recommendation list through the nearest neighbor cluster of users. Experimental results show that the proposed algorithm can effectively accomplish the mission of personalized recommendation and solve the scalability and sparsity problems in userbased collaborative filtering system.