Hybrid Movie Recommendation Based On Deep Neural Networks And Neural Collaborative Filtering

Matrix Factorization and Deep Learning have been applied to the recommendation system and achieved objective results.Matrix Factorization can better extract the implicit features of the user,the recommended object and other latent factors from the user's rating of the recommended object,and match these features to give a recommendation strategy,but this method ignores the explicit features in the existing auxiliary data information of the user and the recommended object.Deep Neural Networks can extracts the features of the user and the recommended object from various auxiliary information,and matches them to give a recommendation strategy,but ignores the implicit features of the user,the recommended object and other latent factors in user rating data.In the recommendation system,only learning the inner product interaction of matrix factorization or using Deep Neural Networks to capture user interaction with item is not enough to effectively learn the features of users and items.To solve this problem,a hybrid movie recommendation model based on Deep Neural Networks and Neural Collaborative Filtering(DNN?HMR)based on explicit feedback and implicit feedback is proposed.The model learns the interaction information between the user and the movie.Firstly,use the basic data of users and movies and the user's rating data of movies to construct a feature matrix of users and movies,and then combine these two feature matrices to build a user-movie feature matrix,and use the Deep Neural Networks to capture the explicit features of users and movies,and obtain the prediction score of the movie recommendation model based on the Deep Neural Networks.Then,on the basis of matrix factorization,the interactive learning ability of Deep Collaborative Filtering is used from the two aspects of implicit mapping and feature mapping to obtain the implicit features of users and movies,and the prediction score of the movie recommendation model based on Neural Collaborative Filtering is obtained.Finally,in the fusion layer,the prediction score of the explicit feature and the prediction score of the implicit feature are linearly combined as the output of the model to obtain the final prediction score.Experiments are performed on the Movie Lens data set,and compared with Generalized Matrix Factorization,Multilayer Perceptron,Neural Matrix Factorization multiple models,and let them compare on Hit Rate(HR)and Normalized Discounted Cumulative Gain(NDCG),experiments show that the hybrid recommendation model can obtain better recommendation performance.