Exploring Explainable Artificial Intelligence Techniques for Hate Speech Detection

Hate speech detection on social media platforms is critical to maintain online safety but poses significant challenges. Traditional machine learning models often fail to capture the nuanced linguistic patterns of hate speech. While deep learning models like LSTM neural networks can effectively model the semantic and syntactic complexity, their lack of interpretability is problematic. This research proposes integrating explainable AI (XAI) techniques like LIME, LRP, Ktrain, and SHAP to enhance deep learning hate speech models’ interpretability. Our key innovation is demonstrating that SHAP provides superior model interpretation, with a score of 0.89, by highlighting discriminatory words aligned with human judgment. The findings show combining deep learning and XAI can advance hate speech detection through interpretable high-performance models. By using SHAP to uncover potential model biases, we can refine systems to be more transparent and fair. This research contributes interpretable hate speech detection advancing online safety and sets the stage for studying model transferability across diverse settings. Overall, our innovative application of XAI to deep learning hate speech detection promotes the development of fair, accountable AI systems to counter online hate.