A novel deep recommend model based on rating matrix and item attributes

Traditional recommendation systems only consider the content of users to predict the rating of items in the recommendation process, and ignore the impact of other factors on the recommendation process except for the user-item rating matrix. Recommendation models that are based on item attributes can portray user preferences and item characteristics from item attribute information, which alleviates the sparseness of rating data to a certain extent. However, they do not consider the potential factors of users and items in the rating matrix. The advantage of deep learning methods in feature representation and feature learning is that they can extract the deep sublinear features of users and items contained in the rating matrix and item attributes. To further improve the quality of recommendation, we proposes the genre rate neural network recommendation (GRNNRec) model, which integrates item attributes based on deep learning. This model integrates the user-item rating matrix and item attributes into a deep neural network to characterize the performance of both in low-dimensional space. First, we use static coding to characterize the properties of the items. Second, we use feature mapping and feature concatenation methods to learn the higher-order features of both continuously. Finally, through the periodic learning rate and the decay rate, we achieve rating prediction. The experiments on different recommendation datasets demonstrate that our model can significantly improve the accuracy of rating prediction in recommendation systems.