Design And Implementation Of Program Recommendation Algorithm Based On Viewing Context Model And Denoise Autoencoder

Compared with traditional TV,the size of smart TV user behavior viewing data has shown an explosive growth.How to dig out the user's preference for programs from a large amount of viewing data has also become the focus of attention,so personalized TV program recommendation the system came into being.This paper establishes a viewing context model that integrates time effects.Through the viewing context model,the scoring matrix is densified and the implicit preference mining is realized.The denoising autoencoder technology is used to perform null value processing and implicit feature interest extraction on the viewing feature vector.Finally,a collaborative filtering program recommendation algorithm based on the viewing context model and denoising autoencoder technology,and taking into account the celebrity effect,is proposed.The main research contents of this subject are as follows:1.Propose a user preference calculation method based on the viewing context model.The explicit scoring matrix of smart TV users for programs is extremely sparse,which cannot well reflect the user's preference for programs.Based on the viewing context information of smart TV users,this paper fully explores the user's implicit feedback on the program from conventional dimensions and time-related dimensions,and combines with the user's explicit feedback to obtain user-program preference that integrates time effects,which can be regarded as a relatively dense user program rating matrix.2.An improved program recommendation algorithm based on denoising autoencoder is proposed.First,the user-program rating matrix still contains some "NULL" values,so this matrix cannot be directly input into the existing denoising autoencoder model.In this paper,a balance matrix is added to the encoding and decoding process of the denoising autoencoder.Through this method,the sparse and null user rating vector is transformed into a more effective and dense hidden layer feature vector.Then,considering that the user's viewing behavior will also be affected by the celebrity effect,this paper proposes the user similarity based on the celebrity effect,and the final user similarity is obtained by fusing the user similarity based on the user feature vector in the collaborative filtering algorithm and Generated a Top-N program recommendation list.The experimental results show that the algorithm in this paper improves the ability of mining unpopular programs while ensuring the accuracy of the recommended results,and verifies the effectiveness and superiority of the algorithm in this paper.