As a method to solve the "information overload" problem brought about by the rapid development of the network, the rec om- mendation system not only saves time and manpower, but also has the appearance of data sparseness and cold start. Deep learning can acquire the in-depth characteristics of users and items, alleviate these shortcomings, and combine with traditional recommenda- tion methods to effectively enhance the recommendation effect. In this paper, the article features are processed by the stack-type noise reduction auto-encoder, the deeper hidden features of the article are extracted, and the recommendation is combined with the probability matrix in the collaborative filtering method, and the data set is used for verification. Experiments show that using a hy- brid recommendation method can improve the efficiency and accuracy of recommendation.