Knowledge transfer for enhanced sentiment-based abusive language detection: Insights from sarcasm detection

Abstract

This paper explores the potential of utilizing sarcasm detection as an auxiliary task to enhance the performance of sentiment-based abusive language detection models. The rise of online platforms has made the detection and prevention of abusive language a pressing issue. However, due to the complex nature of abusive language intertwined with the speaker’s emotion, it remains a challenging task for machine learning models to detect its presence. In this study, we propose a novel approach that leverages the knowledge gained from sarcasm detection to strengthen sentiment-based abusive language detection models. We conduct experiments involving transfer learning and multi-task learning models to compare their performance. Furthermore, we evaluate the robustness and adaptability of the models in a zeroshot setting. The results demonstrate the effectiveness of our approach, with the hybrid model exhibiting superior performance across various metrics. This research contributes to the advancement of knowledge transfer approaches in the field of abusive language detection.