Collaborative filtering recommendation algorithm based on improved denoising auto encoder

Abstract

Aiming at the problems of sparse scoring matrix and low recommendation accuracy of traditional collaborative filtering algorithms, this paper proposes a collaborative filtering recommendation algorithm based on improved denoising auto encoder. First of all, this topic adds a balance matrix to the encoding and decoding process of the denoising auto encoder to compress the high-dimensional and sparse user behavior vector into a low-dimensional and dense user feature vector. Then, the user similarity is calculated in the process, celebrity factors are considered to obtain user similarity based on celebrity effect. Finally, a program recommendation list is generated based on the final user similarity. Experimental results show that the algorithm enhances the performance of scoring prediction, and improves the accuracy and recall rate of recommendation results.