Enhancing Sentiment Analysis with Explainable AI: A Comparative Study of LSTM, GNN, and Capsule Networks

Abstract: Sentiment analysis is vital for applications like social media monitoring and customer feedback analysis. While deep learning models such as Long Short-Term Memory Networks (LSTMs) and Graph Neural Networks (GNNs) achieve high accuracy on large datasets, their lack of interpretability remains a challenge. This paper proposes a hybrid approach combining advanced deep learning models with Explainable AI (XAI) techniques to enhance transparency without sacrificing performance. An LSTM model integrated with SHAP (SHapley Additive exPlanations) achieved 76% accuracy on a dataset of 7,613 tweets, providing insights into word- level contributions to sentiment predictions. Future work includes extending this framework with GNNs and Capsule Networks using XAI tools like GNNExplainer and LIME to capture complex relationships and hierarchical structures, ensuring both accuracy and interpretability for real-world applications.