Singular value decomposition technique has been widely used among the personalized recommendation system. By matrix singular value decomposition can improve the accuracy of personalized recommendations. The traditional model only do the singular value decomposition of the matrix is decomposed into user feature matrix and item feature matrix, and the prediction score is not considered different information containing a different range of scores. By calculating scores critical value, the scoring matrix split into two matrices, called positive feedback matrix and negative feedback matrix. Then two feedback matrices based on the feature to complete the scoring in the prediction. In the experimental data, as used herein MovieLens data sets, the traditional model of singular value decomposition (SVD) and based on hypergraph singular value decomposition model (HSVD) to improve it. Experimental results show that the effects of PSVD, PHSVD model are better than the original model.