Sarcasm Multimodal Dataset

Sarcasm detection is a major challenge in data analysis as it involves complex textual and visual elements. This study builds a multimodal dataset of sarcasm from the Twitter (X) platform, which includes text, images, and a combination of both. The dataset consists of eight main columns, including original text, clean text, text in images, visual elements, and final labels. Labels were manually assigned by four annotators using a binary scheme, with 0 for non-sarcasm and 1 for sarcasm. Dataset validation was conducted using Cohen's Kappa to measure consistency between annotators. The validation results showed an average Kappa value of 0.89, with a high of 0.95 and a low of 0.84, reflecting a substantial to near-perfect level of agreement. This process ensures the reliability of the dataset with minimal noise. This dataset is designed to support the development of machine learning and deep learning-based sarcasm detection models. With a combination of textual and visual elements, and rigorous validation, it offers a solid foundation for the exploration of cross-modality interactions and the development of intelligent applications, such as sentiment analysis and multimodal classification. This dataset contributes significantly to supporting research in the field of sarcasm analysis and artificial intelligence technologies.