A Novel Chinese Sarcasm Detection Model Based on Retrospective Reader

Lei Zhang, Xiaoming Zhao, Xueqiang Song, Yuwei Fang, Dong Li, Haizhou Wang

Abstract: Sarcasm is a subtle form of language in which people express the opposite of what is implied. Existing research works for Chinese sarcasm detection focused on extracting features of target texts. However, there is a lot of contextual information on online social networks, which is insufficient to detect sarcasm based only on target texts. In this paper, we construct a large-scale Chinese sarcasm dataset with contextual information. Meanwhile, a sarcasm detection method based on deep learning is proposed. We used a retrospective reader in the detection process, which includes two parallel modules: Sketchy Reading and Intensive Reading. The Sketchy Reading module reads the target text and contextual information to get an initial impression. The Intensive Reading module uses a hierarchical method to get an intensive impression. Finally, we integrate the two parts to get the final prediction. Evaluation results on the dataset demonstrate the efficacy of our proposed model and the usefulness of contextual information for Chinese sarcasm detection. The research in this paper provides methods and ideas for future work in Chinese sarcasm detection on other social networking platforms.