Research On Hybrid Movie Recommendation Model Based On Deep Learning

Traditional collaborative filtering algorithms still have serious trouble in data sparsity and a large number of noisy data in both feature extraction and similarity calculation,which leads to problems such as low recommendation efficiency and low user satisfaction.Last several years,In pace with the blowout development of Deep Learning&Neural Networks,deep learning can effectively solve various problems in traditional recommendation algorithms.Applying deep learning to recommend field has become one of the most popular topics in current research.Our thesis focuses on the complex feature extraction and high dimensionality and sparsity of traditional collaborative filtering algorithms,and optimizes the auto-encoder neural network in deep learning.It mainly uses MovieLens public movie review data set as the input of the recommendation system,and finally output the recommended list of movies that the user is interested in.The main work is as follows:(1)Aiming at the problems of data sparsity and noisy data in traditional collaborative filtering algorithms,We proposed a recommendation algorithm combining deep denoising autoencoder and attention mechanism.First,the network structure of the traditional autoencoder is improved,and the generalization ability and robustness of the traditional autoencoder are further improved.The user item score matrix is input into the improved deep denoising autoencoder network to learn the nonlinear hidden features of the data.Second,the nonlinear hidden features are integrated into the collaborative filtering recommendation algorithm based on item similarity to mine the linear hidden features of the data.Finally,in order to reflect the difference between different interaction histories,a neural attention mechanism is added to punish the impact of active users.(2)Aiming at the problems that the traditional neural network collaborative filtering model cannot extract the hidden features of the data and the feature data is of high dimensionality and sparseness,a neural collaborative filtering algorithm based on a hybrid deep autoencoder is proposed.First,increase the depth of the traditional autoencoder network,and use a new activation function to propose a hybrid deep autoencoder model to generate deep feature vectors of users and items;secondly,an improved neural network collaborative filtering based on explicit feedback is proposed The model is used to extract the generated deep features of users and items.The model integrates linear and non-linear features of users and items to further enhance the ability of traditional neural network collaborative filtering models to extract features.We compared our two algorithms with several traditional and advanced recommendation algorithms,and the comprehensive performance of the algorithms and the parameters that affect the experimental results are analyzed.The results indicate that our two algorithms are advanced and effective.Applying the two proposed algorithms to the field of movie recommendation can not only bring more economic benefits to service providers,but also bring better user experience to consumers.