Nonlinear Matrix Factorization with BatchNorm

With the exponential increase of information in the high-tech era, how to quickly find information that is valuable to oneself has become a major problem in people's lives. As the main force to solve this problem, the recommendation system can predict the user's rating of the item, and recommend the item with a higher rating to the user. However, due to the increasing amount of various information, the problem of data sparsity has become increasingly prominent, which greatly affects the scoring accuracy and recommendation quality of the recommendation system. In order to solve this problem, this paper integrates Nonlinear Matrix Factorization and batch normalization model BatchNorm to improve the recommendation quality under the condition of data sparsity and increase the training speed. Experiments prove that the algorithm model in this paper has a better recommendation effect.