Deep Sarcasm Detection with Sememe and Syntax Knowledge

Zhiqiang Zhang, Jiajun Shan, Haiyan Wu, Yanhong Chen, Jinxun Jiang, Weijie Wang

Abstract: Sarcasm detection is a challenging task in sentiment computing since sarcasm belongs to a prevalent and sophisticated linguistic phenomenon. Detecting sarcastic sentiment in utterances can avoid misunderstanding the true intentions of speakers and promote the development of natural language processing. However, existing approaches make effort to design complex model structures, ignoring that knowledge is also crucial for sarcasm recognition. In this paper, we propose a novel Sememe and Syntax Knowledge enhanced Sarcasm Detection (SSK-SD) integrating linguistic knowledge (i.e., sememe knowledge and syntax knowledge), where semantic knowledge captures the potential contradictory emotions between words. In contrast, syntactic knowledge enhances the latent semantic representation of the sentences. Experimental results on two benchmark datasets (i.e., Headlines and SARC) demonstrate that SSK-SD significantly improves the state-of-the-art methods on the Sarcasm Detection task, further indicating that knowledge can enhance Sarcasm Detection.