Research And Application Of Hybrid Recommendation System Based On Tensor Decomposition And Deep Learning

In the information society where network services are developing at a high speed.the information data stored on the Internet has been exponentially increasing,causing a rather serious "information overload" problem,which ultimately makes it more difficult for users to obtain the content information efficiently.Therefore,this paper uses the recommendation system to analyze user needs in response to the increasingly severe "information overload" problem.At present,the tensor decomposition algorithm is one of the important methods for recommending system research.The algorithm decomposes the third-order rating data of the user-item-tag into latent factor of users,items,and tags in the implicit factor space.The tensor decomposition algorithm itself has a solid theoretical basis,easy algorithm implementation and accurate prediction.However,the tensor decomposition algorithm still has problems such as poor recommendation of sparse data,single data processing.However the deep learning combines from the low-level features to form high-level abstract features,It can automatically analyzes the hidden features in multi-source heterogeneous data,and solves the problem that the tensor decomposition algorithm performs poorly and the data source structure is single.However,deep learning bases on implicit features and directly analyzes multi-source heterogeneous data.The recommended result is the connection weight between neurons in each layer of deep neural network.It is difficult to give a relatively reasonable explanation.Currently,scoring data is not used as a source of analytical data.Therefore,this paper combines the tensor decomposition method and the deep learning algorithm,integrates the scoring data and the multisource heterogeneous auxiliary data as the analysis data source,enhances the recommendation's personalized precision,and appropriately increases the interpretability of the recommendation results.Finally,through multiple sets of quantitative comparison experiments,the hybrid recommendation algorithm in this paper is more accurate in the social label recommendation system,which is 34.0% less than the traditional collaborative filtering algorithm.This paper focuses on the tensor decomposition algorithm and deep learning,deeply analyzes its own problems,and improves it.It also proposes to combine the tensor decomposition algorithm with the deep learning network to improve the personalized precision of recommendation.The main work and contents of this paper are as follows:First,the optimization and improvement of the tensor decomposition algorithm.Second,the tensor decomposition algorithm and deep learning are combined to improve the recommendation accuracy.Third,the verification and application of the BRATDDL algorithm.In order to investigate the performance of the BRATDDL hybrid recommendation algorithm in practical applications,the application of the online recommendation service of the WEB communication protocol architecture and the improvement of the scientific health motion recommendation system are respectively applied.