Research On Movie Hybrid Recommendation System Based On Deep Learning

The information overload problem has always plagued Internet users.With the advent of recommendation systems,people have a new excellent information filtering tool.However,traditional recommendation algorithms still have many problems,such as cold start and low accuracy.In the past few decades,researchers have been committed to solving these problems and proposed various recommendation models.Therefore,the recommendation system is constantly evolving.This article first introduces the recommendation system,two traditional recommendation models:Content-Based Filtering(CBF)and Collaborative Filtering(CF),and their problems,including cold start,data sparsity,diversity problem,and non-linear feature learning capabilities.However,deep learning-based recommendation systems can cope with these problems and achieve remarkable results.Therefore,two deep learning-based models and a weighted hybrid model are proposed to recommend movies to users.The main research work is as follows:First,this article combines Text Convolutional Neural Network(TextCNN)and Deep Neural Network to propose a Content-Based Filtering(CNN-Based CBF)recommendation model.The model learns the implicit representation of users and movies,then generates a movie recommendation list.Experimental results show that the model still has better performance on data sets with high data sparsity.For Precision,Recall and Mean Square Error(MSE),these three evaluation criteria are better than the traditional recommendation model.Second,based on the Long Short-Term Memory(LSTM)neural network,a Collaborative Filtering(LSTM-Based CF)recommendation model is proposed.The model performs a Matrix Factorization algorithm and decomposes the co-occurrence matrix to obtain the implicit vectors of the users and movies,respectively.The system can use the users and movies feature matrix to make movie recommendations by calculating the Cosine Similarity or predicting the movies' rating scores,and a Top-N list with the highest ratings or similarity is sent to users.Experimental results show that the model is better than the traditional recommendation model and slightly inferior to the CNN-Based CBF model in this article in terms of Precision,Recall and MSE.But its coverage rate is higher than the CNN-Based CBF model.Third,this article combines the above two models in a parallel design to form a weighted hybrid model.The weighting strategy is implemented using a polynomial regression algo-rithm.Experimental results show that this model has a lower MSE.Its recommendation has a high coverage rate,and the results are diverse.At the same time,it can also form a recommendation list for new users and movies without historical rating data.Therefore,the hybrid recommendation system proposed in this article does not have a cold start problem and can perform well on data sets with high sparseness.Its overall performance is better than the other two single models proposed in this article and the widely used deep learning-based models AutoEncoder and DeepFM.In summary,the movie hybrid recommendation system based on deep learning proposed in this article overcomes the shortcomings of traditional recommendation models and improves recommendation accuracy.The model is feasible and effective,and has certain research significance.