Personalized Movie Recommendation based on Convolutional Neural Network

Abstract

In pace with the development of the economy, the spiritual entertainment brought by movies is increasingly valued by people, and the problem of how to recommend the most suitable movie for users among the numerous movies also arises. Based on this, experiments were conducted using convolutional neural networks in the field of deep learning for movie recommendation. The convolutional neural network was trained using user information, movie information, and user movie rating data from the Douban Movie Network. In data preprocessing, instead of converting category fields to one hot encoding, they are converted into numbers and used as indexes for embedding matrices. After extracting features from the embedding layer, input these two features again in the fully connected layer, output a value, and then regress this value to the score to obtain MSE loss, and use the recall rate and NDCG indicator evaluation model. By comparing the accuracy with the other four recommendation algorithms on the same dataset, it was found that the movie recommendation algorithm obtained through the convolutional neural network approach studied in this paper has shown better results on the same evaluation indicators.