<![CDATA[Shopee is a widely used marketplace for online transactions, where product reviews allow users to share their opinions on purchased items. The large volume of reviews makes manual analysis difficult, while previous studies often overlook the normalization of informal language. This study analyzes sentiment in Shopee product reviews using Multinomial Naïve Bayes (MNB) with TF-IDF weighting and custom text normalization, comparatively evaluated against Support Vector Machine (SVM) and validated using 5-fold cross-validation. The dataset consists of 500 balanced reviews (250 positive and 250 negative) obtained from Kaggle. Preprocessing includes case folding, cleansing, normalization, tokenization, stopword removal, and stemming. Hold-out evaluation (80:20) shows that SVM achieves 96.00% accuracy, 92.59% precision, 100.00% recall, and a 96.15% F1-score, while MNB achieves 91.00% accuracy, 84.75% precision, 100.00% recall, and a 91.74% F1-score. Five-fold cross-validation yields mean accuracies of 90.80% (±3.25%) for MNB and 89.60% (±2.50%) for SVM. These results indicate that TF-IDF with custom text normalization provides stable classification performance for e-commerce reviews.]]>
Copyrights © 2026
