In the digital economy era, e-commerce platforms like Shopee receive thousands of user reviews daily, which significantly influence customer perceptions and purchasing decisions. However, sentiment analysis of such reviews remains challenging due to the presence of noise, uncertainty, and dynamic data changes. This quantitative research aims to develop a more reliable sentiment classification model by integrating a Lexicon-Based labeling approach and Support Vector Machine (SVM) classification with a Robust Optimization framework. The labeling process uses a sentiment lexicon dictionary that assigns polarity values to words, classifying texts into positive, negative, or neutral categories. The classification process utilizes SVM to evaluate sentiment prediction based on key performance metrics: Accuracy, Precision, Recall, and F1-score. These performance metrics are treated as uncertain parameters in the optimization phase. The main contribution of this study is the formulation of a robust optimization model for sentiment analysis weighting problems, transforming a multi-criteria objective into a single-objective utility function. By applying polyhedral uncertainty modeling, the robust counterpart formulation accounts for worst-case scenarios in model evaluation. Numerical experiments using Python in Google Colab show that while the deterministic model achieves a higher performance score (0.865), the robust model yields a slightly lower score (0.825) but offers better stability under uncertainty. These results imply that robust optimization can enhance the reliability of sentiment classification systems in real-world e-commerce applications, providing more trustworthy insights for businesses in managing consumer feedback.
Dongoran et al. (2025) studied this question.