Sarcastic RoBERTa: A RoBERTa-Based Deep Neural Network Detecting Sarcasm on Twitter

Maciej Hercog, Piotr Jaronski, Jan Kolanowski, Pawel Mieczynski, Dawid Wisniewski, Jedrzej Potoniec

Abstract: Sarcastic RoBERTa is an approach to recognizing sarcastic tweets written in English. It is based on a pre-trained RoBERTa model supported by a 3-layer feed-forward fully-connected neural network. It establishes a new state-of-the-art result on the iSarcasm dataset, attaining the \(F_1\) score of 0.526, and being not far from the human performance of 0.616.