Sarcasm Detection in Tunisian Social Media Comments: Case of COVID-19

Abstract

The goal of this study is to learn more about how the public sees COVID-19 pandemic behaviors and to identify important themes of concern expressed by Tunisian social media users during the epidemic. Around 23K comments were collected, written in both Arabic and Latin characters in the Tunisian dialect. Native language experts manually tagged these comments for sarcasm identification (sarcastic and non-sarcastic). In addition to health, our dataset contains comments on entertainment, social, sports, religion, and politics, all of which are impacted by COVID-19. This research examines the sarcasm expressed in Tunisian social media comments regarding the novel COVID-19 from its appearance in the first half of 2020. We also provide benchmarking findings applying machine learning and deep learning algorithms for sarcasm detection. We obtained an accuracy of above 80%.