Interpretable Sexism Detection with Explainable Transformers

With the widespread growth of social media platforms, instances of racism, cyberbullying, and the use of offensive language have surged. Consequently, women face challenges stemming from the presence of sexist content, which not only impedes their self-improvement but also exacerbates feelings of anxiety. Recognizing online sexism as a harmful phenomenon, the need for an automated tool to detect it has become paramount. This paper proposes an automated framework for extracting insights and identifying sexist language with high accuracy, utilizing machine learning (ML), deep learning (DL), and transformer-based models. Then, we incorporate the explainable AI (XAI) technique to enhance interpretability and make it more understandable to humans. To assess the performance of our method, we conducted experiments using a publicly available dataset focused on sexism. The experimental results underscore the effectiveness of our approach in detecting online sexism, surpassing the performance of several state-of-the-art methods.