Few-shot learning for short text classification

Due to the limited length and freely constructed sentence structures, it is a difficult classification task for short text classification. In this paper, a short text classification framework based on Siamese CNNs and few-shot learning is proposed. The Siamese CNNs will learn the discriminative text encoding so as to help classifiers distinguish those obscure or informal sentence. The different sentence structures and different descriptions of a topic are viewed as 'prototypes', which will be learned by few-shot learning strategy to improve the classifier's generalization. Our experimental results show that the proposed framework leads to better results in accuracies on twitter classifications and outperforms some popular traditional text classification methods and a few deep network approaches.