CuSNMF: A Sparse Non-Negative Matrix Factorization Approach for Large-Scale Collaborative Filtering Recommender Systems on Multi-GPU

Abstract

Matrix factorization (MF), as one of the most accurate and scalable approaches in dimension reduction techniques, has become popular in the collaborative filter- ing (CF) recommender systems communities. Currently, Non- negative Matrix Factorization (NMF) is one of the most famous approaches for MF, due to its representative non-negativity fea- ture for CF model. However, it is non-trivial to obtain high per- formance of sparse NMF (SNMF) on Graphic Processing Units (GPU) for large-scale problems, due to the redundant large- scale intermediate data, frequent matrices manipulation, and access on the sparse rating matrix with irregular distribution non-zero entries. In this work, we propose single-thread- based SNMF, which depends on the involved feature tuples multiplication and summation, and then, we present L2 norm regularized single-thread-based SNMF. On that basis, a novel CUDA parallelization NMF approach (CuSNMF) is presented for GPU computing. Furthermore, to process large-scale CF data sets and make advantages of GPU computation power, we propose multi-GPU CuSNMF (MCuSNMF). Compared with state-of-the-art parallel algorithms, CCD++, and CUMF, MCuSNMF obtains the highest performance.