The Sarcasm Spectrum: Comparative Insights into Multimodal Detection

Sentiment analysis, or SA, has developed into a crucial field of study. This has been especially evident in the post COVID-19 period; as social media has taken centre stage as a means of communication. The complicated interplay between these relationships and involvement in virtual discourse spaces, along with textual and multimodal information, introduces severe complications in SA. The subtle identification of irony, humour, sarcasm, negation, and the distinction of sentiment at an Aspect-Level (ASA) are essential to understanding these difficulties. More and more recent research efforts focus on the complex task of sarcasm detection, realizing that it can significantly improve the accuracy and effectiveness of SA models. Sarcasm is a complex linguistic construct that hides negative feelings behind seemingly positive language, making interpretation difficult. It is inherently difficult for computational systems and human mind to identify, needing deep understanding of the context, nonverbal clues such kinesics, prosody, and paralanguage, as well as background knowledge on sociocultural issues. This work aims to provide a thorough discussion on multimodal data integration while critically examining current approaches in sarcasm detection. It also emphasizes how these approaches will impact the development of technologies like intelligent conversational agents, workplace stress analytics, and mental health diagnostics.