Interpretable sentiment analysis of Bangla e-commerce comments using explainable AI techniques

Abstract

This research focuses on sentiment analysis of Bengali e-commerce comments using various machine learning algorithms. The primary objective is to develop and evaluate models that can accurately classify comments into positive or negative sentiments. I experimented with LR, DT, RF, MNB, KNN, SVM, and SGD. My dataset, collected through web scraping, underwent extensive preprocessing to ensure quality and relevance. The performance of each model was assessed using metrics such as accuracy, precision, recall, and F1-score. Among all the models, SVM exhibited the best performance, achieving a training accuracy of 89.6%, validation accuracy of 87.6%, and a model accuracy of 86.7%. This superior performance indicates SVM's robustness and effectiveness in handling high-dimensional data and non-linear decision boundaries in the context of sentiment analysis for Bengali text. In addition to model training and evaluation, I implemented a web application for practical deployment, designed using HTML, CSS, JavaScript, and the Flask framework. This application facilitates user-friendly interaction and real-time sentiment analysis, ensuring scalability and reliability. The integration of Explainable AI (XAI) techniques further enhances the interpretability of the model's predictions, providing insights into the factors influencing sentiment classification. Overall, my study demonstrates the potential of SVM in achieving high accuracy and reliability in sentiment analysis of Bengali e-commerce comments, paving the way for more advanced applications in natural language processing for underrepresented languages.