Abstract: Text classification is the classical research direction in NLP and plays an important role in information processing. At present, deep learning network has achieved the remarkable performance in image recognition, machine translation and other fields and it also has been proved to be capable of learning higher-level sentences and document representation in NLP tasks. In this paper, based on GRU model and the convolutional layer in CNN, we propose a novel hybrid text classification model called Attention-based C-GRU. Moreover, we introduce Attention model in our model, which effectively highlights the role of key words and optimizing the extraction of features. We leverage the model to learn the meaning of text and evaluate it on topic classification, question classification and sentiment classification tasks. The experiment demonstrates the effectiveness of our approach in comparison with baseline models and state-of-art methods.

Key words: text classification, deep learning, Attention model