Text classification of Chinese news based on multi-scale CNN and LSTM hybrid model

Deep neural network has significant performance in text classification. Convolutional neural network (CNN) and recurrent neural network (RNN) are two main structures for natural language processing. They use different ways to understand natural language. In our work, we use the advantages of these two frameworks to propose a hybrid model of multi-scale CNN and Long Short-Term Memory (LSTM). Firstly, we use multi-scale CNN to obtain the features of text sentences, and use LSTM model to capture the dependency of text context. Then the feature vectors generated by the two parts are fused to form a new feature vector, our model has the advantages of CNN and LSTM. Finally, the softmax layer is used for classification. We evaluate the performance of the proposed model in text classification tasks. The results show that the classification performance of our proposed model is better than the traditional classification models, CNN and LSTM, indicating that the classification effect of this model is more significant.