For the traditional matrix factorization algorithm, only the scoring information is used as the recommendation basis, when the scoring data is sparse, the implicit factor vector cannot be accurately extracted, making full use of auxiliary information has become one of the research hotspot, and a recommendation based on deep learning is proposed, the model HRS-DC uses the deep neural network and the convolutional neural network to extract the recessive feature vectors of the user and the project from the auxiliary information, then transforms the feature vector through the improved neural collaborative filtering to obtain a new scoring matrix. Through verification on three real data sets, the accuracy of scoring prediction is improved compared with Probability Matrix Factorization（PMF）, Collaborative filtering Topic Regression（CTR）, Collaborative filtering Deep Learning（CDL） and Convolution Matrix Factorization（ConvMF）, and the cold start problem is alleviated to some extent.

%U http://cea.ceaj.org/EN/10.3778/j.issn.1002-8331.1903-0362