A large corpus for sarcasm research and for training and evaluating systems for sarcasm detection is presented. The corpus comprises 1.3 million sarcastic statements, a quantity that is tenfold more substantial than any preceding dataset, and includes many more instances of non-sarcastic statements. This allows for learning in both balanced and unbalanced label regimes. Each statement is self-annotated; that is to say, sarcasm is labeled by… See the full description on the dataset page: https://huggingface.co/datasets/CreativeLang/SARC_Sarcasm.
