Multimodal Sarcasm Detection (MSD) in Videos using Deep Learning Models

Abstract

Every day, individuals all over the globe use video-sharing websites to broadcast their thoughts, experiences, and recommendations to the world. There has been a huge rise in educational interest in the area of sarcasm detection for these opinionated videos. Although sarcasm has proven effective for written text, it remains an unexplored area of research when applied to video and other forms of multimodal data. Numerous verbal and nonverbal signs, such as a shift in voice, an overemphasis in a phrase, a stretched pronunciation, or a stiff-looking face, are often used to convey sarcasm. The majority of current research on sarcasm detection has heavily relied on textual content. Hence, in this paper, we suggest that the use of multiple modalities can help with more accurate sarcasm identification. The results of our studies on one of the multimodal sarcasm detection (MSD) benchmark datasets “MUSTARD” reveal that our strategy outperforms other algorithms in sarcasm classification.