Enhancing Hate Speech Detection through Explainable AI
Dipti Mittal, Harmeet Singh
Abstract
The potential of XAI in detecting hate speech using deep learning models is versatile and multifaceted. To better understand the decision-making process of complex AI models, this study applied XAI to the dataset and investigated the interpretability and explanation of their decisions. The data was preprocessed by cleaning, tokenizing, lemmatizing, and removing inconsistencies in tweets. Simplification of categorical variables was also performed during training. Exploratory data analysis was conducted to identify patterns and insights in the dataset. The study used a set of existing models, including LIME, SHAP, XGBoost, and KTrain, to analyze the accuracy. The KTrain model achieved the highest accuracy and lowest loss among the variants developed to increase explainability.