Automatic document classification using convolutional neural network

Abstract

Official document classification is an integral essential part of daily archives administration. The traditional manual document classification method is time-consuming and labour-intensive, and the classification effect cannot be fully guaranteed. With the popularity of computers and the development of machine learning, the convolutional neural network model is becoming more and more mature, and the CNN model is suitable for the problems encountered in the current official document classification. This paper proposed a model based on convolutional neural network to solve the problem of official document classification. To train and test the model, we manually classify the dataset into ten categories according to the classification of college archives entities. And it was trained on a dataset with size of 6765. And on the testing dataset with size of 676, it reached an accuracy of 90%. And for comparison, we also trained a LSTM model and a GRU model (they are both popular in the natural language processing field), and results showed that the automatic official document classification method based on convolutional neural network can improve the efficiency of traditional official document classification. Also, it provides a new way of thinking for the solution of the official document classification problem.