Recommended Method based on Noise Reduction Self Encoder

Abstract

The user and item features extracted by the traditional recommendation system are relatively simple, and at the same time, the mining of the internal relationship between users and products is not sufficient, resulting in unsatisfactory recommendation effect. Based on the research of the existing recommendation system, aiming at the problem of insufficient user and item information extraction, this paper realizes the screening and classification of items and users through algorithms, and performs shallow semantic information and deep semantic information on the information of users and items. Divide and process them separately to achieve the purpose of fully extracting shallow semantic features and deep semantic features, and recommend different items to different users according to users' ratings and interests. A recommendation method based on deep noise reduction autoencoder proposed in this paper firstly needs to divide the information of users and items into shallow semantic information and deep semantic information, and process them separately. The autoencoders are trained separately to realize the recommendation of different items for users with different interests. Through the above methods, this paper mainly improves the recommendation accuracy of the current recommendation system and reduces the adverse effects of matrix sparsity.