Explainable AI Models for Decoding Emotional Subtexts on Social Media

Abstract

Social media platforms, such as X (formerly Twitter), provide users with concise but impactful tools to express their views and feelings. Users present their views and express their feelings in hashtags and emojis on a wide range of topics. The sheer volume of this textual data offers a rich source for analyzing public sentiment and emotions. Numerous machine learning and deep learning approaches have been presented lately for optimal emotion detection and sentiment analysis of these tweets. Given the complexity of processing human language, natural language processing (NLP) techniques face the challenge of explainability in their decision-making process. To bridge this gap, we introduce an explainable NLP-based framework for the recognition of human emotions within textual data. We propose a novel recurrent neural network architecture incorporating a bidirectional long short-term memory layer for emotion prediction and sentiment analysis on English tweets. The performance of the proposed model is evaluated with real-world X data against benchmark techniques. The proposed model achieves accuracy, precision, recall, and an F1-score of over 90%, which is higher than the considered benchmark models. Subsequently, we integrate the explainable artificial intelligence (XAI) approaches, namely, local interpretable model-agnostic explanations (LIME) and SHapely Additive exPlanation (SHAP) to explain the decision-making process behind the proposed model’s prediction. Applying these XAI techniques not only boosts the proposed model’s transparency but also reinforces its reliability in accurately processing and explaining textual data.