Sarcastic comments on Reddit and Twitter

Abstract

he paper deals with automatic detection of sarcasm on social media (Reddit and Twitter). The objective of this paper is to explore various expressions used to encode sarcasm in comments tagged as sarcastic by social media users. The difference between sarcasm and irony and the descriptions of these terms are presented in the theoretical part. The datasets extracted from Reddit and Twitter exceeded 1.2 million comments. The research methods used in the study involved Machine Learning (ML) approaches and sophisticated deep learning algorithms (BERT, RoBERTa). It has been observed that: (i) a list of words (mostly pre-modifying amplifiers) can be identified which typically cooccur with sarcastic comments; (ii) the  exclamation mark often accompanies sarcasm; (iii) sarcastic comments tend to have a negative sentiment analysis; (iv) RoBERTa models far outperformed the BERT model in detection accuracy of sarcastic posts.