Parallel Factorization Machine Recommended Algorithm Based on Map Reduce

Factorization Machines [1, 2] is a new factorization model that can combine the merits of SVM model with matrix factorization models. It can model all the interactive actions using factorized parameters. So it could mimic most other matrix factorization models by feature engineering. Due to the superior flexible, Factorization Machines has already been widely used in many recommended algorithm competitions and practical online recommended system. But, because of the prevalence of large dataset, there is a need to improve the scalability of computation in factorization machines model. In this paper, we propose a parallel algorithm can be used on Factorization Machines model. The experimental results show that the proposed algorithm has good speed-up and scalability on big dataset.