Research And Application Of Recommendation Algorithm Based On Tensor And Matrix Completion

With the rapid development of Internet information technology,the proble m of information overload has become more and more serious.Users cannot q uickly find information with high value and satisfaction from massive data,so the auxiliary technology recommendation system came into being.Nowadays,m any Internet platforms such as e-commerce,social news,and Internet audio are using recommendation algorithms to improve the intelligence of the system.T he recommendation system has gradually become an indispensable part of Inter net applications.This paper is based on the basic principle and working proces s of recommendation system,this paper proposes different algorithms to solve t he data sparsity problem of recommendation system,mainly focusing on tensor Completion technology,matrix Completion technology and transfer learning tech nology.direction.The main work of this paper is summarized as follows:(1)Propose a probability-based CP tensor Completion recommendation alg orithm.Firstly,according to the shortcomings of the existing CP Completion al gorithm,in the original CP decomposition process,the Gaussian matrix and the Gaussian tensor are used to simulate the correlated noise data and missing dat a,and then the factor matrix is updated and trained by the minimum alternatin g square method,and finally a more complete The approximate tensor realizes the prediction Completion of the original data,and the new data constructed ca n better reflect the characteristics of the original data,and the algorithm has hi gher robustness and computational advantages.(2)A FunkSVD-based implicit data transfer learning recommendation algor ithm(FunkSVD Transfer Learning)is proposed.In this paper,based on the ch aracteristics of implicit data,this paper models and analyzes it,constructs a us er implicit click data matrix,and fills it with FunkSVD algorithm,then migrat es the filled click data to obtain a user indirect similarity feature.It is combin ed with the user direct similarity feature obtained in explicit data to improve t he accuracy of similarity calculation.Experiments show that FunkSVD-TL(Funk SVD Transfer Learning)provides more suitable data for subsequent calculation.