Hybrid Collaborative Filtering Algorithm Based On Deep Learning

With the application of the Internet and the popularization of information technology,the entire society has entered an era of data explosion and information redundancy.Massive unstructured information is overloaded,so users often need to consume a lot of time and energy to find out what they need.The recommendation system is an indispensable and important tool to solve this problem.Nowadays,there are three main types of recommendation systems: one is content-based recommendation,the other is collaborative filtering recommendation,and the third is hybrid recommendation.Content-based recommendations can accurately learn user characteristics and are highly interpretable,but face limited analysis content,excessive specialization,and cold start issues;collaborative filtering does not require too much expertise in the corresponding data field,but it still cannot effectively avoid user cold Start-up,data sparsity and other issues;the hybrid recommendation system is based on a variety of different recommendation technologies,integrated to achieve their complementary advantages to achieve better results.The main work of this article is as follows:1)Aiming at the problem of the cold start of the recommendation system and the slow speed of the traditional recommendation algorithm,a matrix decomposition model with increments is proposed.Based on the matrix decomposition algorithm,the model incorporates the paranoid factors of users and projects.In the iterative calculation process,the batch learning algorithm with increments is used to accelerate the training process.Experiments based on the datasets Movie Lens-100 k and Movie Lens-1M show that compared with the traditional matrix factorization algorithm,the root mean square error(RMSE)of the matrix factorization model with increment is reduced by about 0.3compared to the matrix factorization algorithm %-0.7%,about 18%-21% increase in training efficiency;2)Aiming at the problem of cold start and sparsity of the recommendation system,this paper proposes a recommendation model that combines long-short-term memory network and probability matrix decomposition,and a deep probability matrix decomposition model.The model decomposes the user-item scoring matrix based on the probability matrix decomposition structure to obtain user characteristics;and uses the long-term and short-term memory network's powerful learning capabilities to deeply mine project auxiliary information to obtain project characteristics;finally,the two are organically combined to effectively improve the algorithm Performance.Experiments show that the RMSE of this model is effectively reduced by 2%-4% compared to other recommended models based on deep learning;3)Integrate the matrix decomposition model with increment and the depth probability matrix decomposition model to propose the depth probability matrix decomposition model with increment.Based on the complementary advantages of the two models,the model uses the former to improve the learning efficiency of the algorithm,and the latter to improve the learning ability,fitting ability and anti-interference ability of the algorithm.The experiment proves that the RMSE of the deep collaborative filtering model with increment is reduced by about 1%-7% compared with other algorithms,and the training efficiency is improved by about 15%-25%.