A Multimodal Framework For Satire Vs. Sarcasm Detection

Sarcasm and satire are common ways to communicate complicated ideas with humor and irony, both expressing criticism or mockery. Detecting sarcasm vs. satire in real life can foster better communication by understanding intent, preventing misunderstandings, and promoting critical thinking and empathy. We introduce an innovative dataset for detecting Multi-modal Satire vs. Sarcasm, leveraging the MUStARD dataset for its construction. This dataset includes visual utterances annotated with sarcasm and satire labels, enriching its potential for studying multimodal satire and sarcasm detection. This enriched dataset serves as a robust foundation for training and evaluating machine learning and deep learning architectures that combine textual and visual cues to identify satire and sarcasm in multimedia content. By incorporating this enhanced dataset, the research aims to advance the field of multimodal satire and sarcasm detection and contribute to a deeper understanding of humor and irony in online communication. We carry out experimentation with various architectures like LSTM, GPT2, BERT, SVM, and Logistic Regression. The proposed architecture achieves an F1 score of $\mathbf{0. 9 5 6}$, outperforming other baselines.