Research On Online Rating Prediction In Video Personalized Recommendation System

Nowadays,all major video websites have personalized recommendation sections,from Hulu and Netflix in foreign countries to IQIYI and Tencent in China.For users,personalized recommendation of videos can alleviate the problem of information overload in video websites,and enable users to quickly access to the content they are really interested in,so as to facilitate user operation and improve user experience.For video sites,a good performance of the recommendation system can increase the user viscosity,improve the economic benefits of the site.So in the video website a recommendation system is very necessary.In the personalized video recommendation system,the online rating prediction of the video is an important part.One of the most popular and commonly used online scoring prediction algorithms is the Matrix Factorization algorithm in Collaborative Filtering.Its advantages are more intuitive and efficient,so this paper will use the matrix decomposition algorithm carry out the research.However,Matrix Factorization only uses single information of scoring matrix for mining and learning,will recommend precision caused by scoring matrix data sparse problem such as unable to further improve,so this article will be on the basis of the evaluation matrix,the introduction of video auxiliary information to make data sparse this up,and choose the depth study of stack denoising autoencoder(SDAE),using its characterization of strong learning ability to extract video auxiliary information hidden in the factor,then SDAE learned from the auxiliary information implied characteristics into Matrix Factorization algorithm,Through more dimensions and deeper level of information and data learning,the accuracy of score prediction can be improved.Therefore,this paper proposes a collaborative filtering rating prediction hybrid model based on deep learning to predict online ratings in personalized video recommendation systems.This paper mainly carries out the following work:1.Establish a collaborative filtering score prediction hybrid model based on deep learning.Firstly,the deep learning algorithm is used to SDAE,which can effectively learn the auxiliary information of the project.The hidden feature matrix of the video is extracted through the deep training of stacking de-noising machine for many times.Secondly the collaborative filtering algorithm matrix decomposition algorithm is selected to learn the intrinsic characteristics of user-project scoring matrix;Next,the implicit feature matrix of the project extracted from deep learning is integrated into the Matrix Factorization,then the loss function of the hybrid model is established,and the stochastic gradient descent method is used for optimization,and a new high-precision prediction score matrix is obtained through training.2.Optimized design.The user implicit feedback part of matrix decomposition is trained by local optimization to reduce computation and improve the overall performance.Moreover,the particle swarm optimization method is used to search and optimize the size of partial super parameters in deep learning,so that the results of the hybrid model are better and the score prediction effect is more accurate.3.Experimental verification and analysis.The influence of SDAE tests with different levels on the hybrid model;The performance of particle swarm optimization for hybrid model was verified.Compare the hybrid model with the baseline model.Experimental results show that t the learning effect of the mixed model depends more on the learning effect of a certain data information in SDAE than on the design of layers.The particle swarm optimization(PSO)algorithm has better performance on the small scale data set.The mixed model was compared with the baseline model and other deep learning-based scoring prediction models.The experimental results show that the learning effect of the mixed model depends more on the learning effect of SDAE for a certain data information than on the design of layers.The particle swarm optimization algorithm(PSO)performs better on the small data set for the super parameter optimization of the mixed model.In the comparison experiment,the experimental results of the hybrid model are better than other model,which proves that the collaborative filtering score prediction hybrid model based on deep learning performs well and can further improve the accuracy of the score prediction.