An Interpretable and Optimized Attention-Driven LSTM Framework for Email Sentiment Classification

Abstract

The analysis of email sentiment is an important aspect of natural language processing because it allows the automated detection of user sentiments. Existing machine learning approaches tend to focus on incorporating dependencies between words and their significance among texts under email analysis. This study offers an innovative solution based on deep learning using Long Short-Term Memory (LSTM) model that includes an attention mechanism. The proposed framework incorporates an LSTM unit to retain the long term dependencies in the email text and an attention mechanism that focuses on the most relevant words for efficient sentiment classification. Hyperparameter tuning is performed for the proposed model via the Bayesian optimization method for better convergence and generalization. For the experimental evaluations, various baseline machine learning and deep learning methods are applied and their performances are compared with the attention-based deep learning models. From the experimental findings, the proposed Attention-based LSTM model outperformed the other models with an accuracy of 98.63%, F1-Score of 90.39%, and Area Under the Curve (AUC) of 0.99. It is also confirmed from the comparison study that the use of deep learning models is better than other techniques in terms of understanding sequential dependencies in the input data. Further improvement in the accuracy of the classification results is achieved through the application of an attention-based technique that emphasizes informative aspects. The incorporation of explainable AI methods into the framework provides valid explanations for model predictions, interpretability and reliability. This study reveals the effectiveness of attention-enhanced sequence models for robust and scalable email sentiment classification.