Sarc7: Evaluating Sarcasm Detection and Generation with Seven Types and Emotion-Informed Techniques

The paper introduces Sarc7, a benchmark for classifying seven types of sarcasm—self-deprecating, brooding, deadpan, polite, obnoxious, raging, and manic—using the MUStARD dataset. Classification was tested with various techniques, including zero-shot, few-shot, chain-of-thought, and a novel emotion-based prompting method, with Gemini 2.5 achieving the highest performance at an F1 score of 0.3664. This work is significant for AI practitioners as it enhances sarcasm detection and generation capabilities in large language models, addressing the complexities of human communication.