Reproduce “A Survey of Multimodal Sarcasm Detection” on OpenResearch

About this paper

Shafkat Farabi, Tharindu Ranasinghe, Diptesh Kanojia, Yu Kong, Marcos Zampieri

This paper presents the first comprehensive survey on multimodal sarcasm detection (MSD), consolidating research from 2018 to 2023 across visuo-textual and audio-visual-textual modalities, outlining existing datasets and computational models, including deep learning, multimodal transformers, and early LLM-based approaches.

Key insights

• Multimodal cues are indispensable for robust sarcasm detection, as incongruity between modalities often signals sarcastic intent that is not discernible from text alone.
• The field has evolved from simple feature concatenation in traditional deep learning to sophisticated multimodal transformers and initial explorations with Large Language Models (LLMs).
• There is a current lack of multilingual datasets, limited consideration of perspectivism in annotation, and a need for exploring inter-task dependencies with other non-literal language forms.

Summary by alphaXiv.