Text Classification with Keywords and Co-occurred Words in Two-stream Neural Network

In the task of text classification, the results are often affected by the sparseness of the corpus in a certain category. This paper proposed a Text classification method based on keywords and co-occurred words in two-stream neural network, in which the keywords are selected from training corpus, and the co-occurred words are chosen from external corpus, and they both are the input of the aid-stream of neural network, which is to enhance the category characteristics of representation vector in main-stream, especially for the category with less data distribution. The experiment results in Fudan University Corpus proved that the proposed method further improved the accuracy of the traditional GRU network. The proposed aid-stream based on keywords and co-occurred words is of great significance to the task of text classification and will provide method support for other natural language processing task.