Multi-modal Sarcasm Detection Based on Contrastive Attention Mechanism

In the past decade, sarcasm detection has been intensively conducted in a textual scenario. With popularization of video communication, analysis multi-modal scenarios received much attention recent years. Therefore, detection, which aims at detecting conversations, becomes increasingly hot both natural language processing community and community. this paper, considering that is often conveyed through incongruity between modalities (e.g., text expressing compliment while acoustic tone indicating grumble), we construct Contrastive-Attention-based Sarcasm Detection (ConAttSD) model, uses an inter-modality contrastive mechanism to extract several features for utterance. A feature represents information two modalities. Our experiments on MUStARD, benchmark dataset, demonstrate effectiveness proposed ConAttSD model.