A Human-Centric Evaluation of XAI Methods for Sentiment Analysis on Amazon Reviews

This study presents a human-centric evaluation of Explainable Artificial Intelligence (XAI) methods, specifically LIME and SHAP, applied to sentiment analysis of Amazon product reviews. Our objective was to improve interpretability and user trust in AI systems by combining quantitative model evaluation with qualitative insights into user perception. Using a Logistic Regression model for sentiment classification, we apply XAI techniques to explain individual predictions and analyze the contribution of input features. The empirical results are evaluated using accuracy, F1 score, precision, recall, and ROC-AUC, whereas human-centric analysis focuses on trust, usability, and transparency. Our findings highlight both the promise and limitations of current XAI approaches in enhancing transparency and fostering responsible AI adoption.