Performance comparison of Korean keyword-based document classifiers using convolutional neural networks

Recently, with the rapid development of deep neural network (DNN) technology, application research is increasing in various fields of artificial intelligence. Particularly, activity is increasing in the image recognition field using DNN. Deep learning-based technology is also used in document classification, having been studied in the natural language processing field for a long time. Document classification using sentence unit texts has been attempted in other countries with high performance. In this study, document classification is performed based on an existing convolutional neural network (CNN) containing Korean/English keyword data from Korean reports; the results are summarized. A singlelayer CNN was used. Owing to the experimentation with bigram, we obtained 24% more accurate results than simple word-based learning using Korean learning data.