Abstract Matrix factorization algorithm is one of the recommendable algorithms. In order to tackle the inefficiencies of the traditional matrix factorization algorithm like the slow training time and the insufficient computing resource for the mass data, a parallelization algorithm of singular value decomposition (SVD) under the Spark framework is proposed to perform SVD, standardization, and dimensionality reduction for the user-rating matrix, and obtain the user-feature matrix and project-feature matrix. The recommendation model is obtained by determining the prediction rating. MovieLens data show that this algorithm can significantly shorten the training time of the model, improve the running efficiency of the recommendation algorithms for the mass data, and improve the algorithm accuracy.