With the rapid development of artificial intelligence, the automatic classification of literature is changing from the rule-based to the machine learning. After an outline of deep learning, the paper introduced convolution neural network into the automatic classification, constructing a multi-level model based on the title and the key words and thus CLC is given automatically. Through the deep learning model in TensorFlow, about 1700000 records of National Newspaper Index were used to make model train. More than 7000literature were processed with the model and the result is: under the production condition, the accuracy of the first classification is 75.39%; the accuracy of the fourth classification is 57.61. When the confidence is 0.9, the correct rate of the first classification is 43.98%, error rate is 1.96%; correct rate of the fourth classification is 25.66%, the error rate is 5.11%.This shows that the model can be used to help realize the semi-automatic in the classification of National Newspaper Index and other problems.