Research On Personalized Recommendation Algorithm Based On Deep Learning And Matrix Factorization

As an effective information filtering technology,personalized recommendation can effectively alleviate the problem of information overload,and has strong practical value and commercial value.Due to the complexity of the business environment faced,personalized recommendations have always had problems such as data sparseness,cold start,and user interest migration that need to be resolved and optimized.In this paper,the existing matrix decomposition recommendation algorithm is improved according to the data sparseness and user interest migration in the recommendation,and on the basis of it,a deep learning and matrix decomposition related recommendation model is proposed,and a recommendation algorithm based on deep learning and matrix decomposition is proposed.,Mainly doing the following work:First,according to the asymmetric relationship between different users and different items,an improved correlation calculation formula is used to predict the score,and a matrix decomposition recommendation algorithm based on asymmetric similarity is proposed to alleviate data sparseness caused by asymmetric user information problem.And on the MovieLens,Amazon and Ciao data sets have been experimentally verified,the experimental results show that compared with the same type of recommendation algorithm improved algorithm can significantly improve the algorithm's recommendation accuracy.Secondly,considering that the neural network can learn more flexible mapping relationships from the data,on the basis of the improved matrix decomposition recommendation algorithm,a multi-layer perceptron is added to obtain the nonlinear part of the user-data interaction information,and a weighted neural network is proposed.The network matrix factorization recommendation algorithm has been experimentally verified on the MovieLens and Pinterest datasets.The experimental results show that compared with other mainstream recommendation algorithm models,the recommendation accuracy has also been improved to a certain extent.In addition,for the problem of user interest changing with time,the recurrent neural network GRU and the weighted neural network matrix decomposition recommendation algorithm are combined.The weighted neural network matrix decomposition recommendation algorithm is used to predict user long-term interest,and the GRU model is used to predict user short-term interest.A weighted neural network matrix decomposition recommendationalgorithm fused with GRU,and experimentally verified on the two data sets of MovieLens and Pinterest.Compared with the weighted neural network matrix decomposition recommendation algorithm,the weighted neural network matrix decomposition recommendation algorithm fused with GRU is more accurate There has been a significant improvement,and compared with other mainstream recommendation algorithms,its recommendation accuracy also performs better.Finally,the demand analysis of the personalized movie recommendation system is carried out,and the personalized movie recommendation system is designed and implemented by using the movie data provided by TMDB according to the matrix decomposition recommendation algorithm based on deep learning proposed in this paper.