<![CDATA[The rapid growth of the skincare industry has generated a massive volume of consumer reviews on e-commerce platforms, making manual sentiment analysis increasingly impractical. This study compares the performance of Multinomial Naïve Bayes and Support Vector Machine (SVM) for sentiment classification of skincare product reviews using the Sephora Products and Skincare Reviews dataset from Kaggle, consisting of 602,130 reviews. Unlike previous studies that often employ different datasets and experimental settings, this research evaluates both algorithms using a uniform pipeline, including automatic sentiment labeling based on rating values, text preprocessing, Bag of Words feature representation, and identical evaluation procedures. Model performance was assessed using Area Under Curve (AUC), Accuracy, Precision, Recall, F1-Score, and Matthews Correlation Coefficient (MCC). The results indicate that positive sentiment dominates the dataset (82.37%), resulting in a highly imbalanced class distribution. Multinomial Naïve Bayes achieved better performance than SVM on most evaluation metrics, with an AUC of 0.784, F1-Score of 0.764, Precision of 0.749, and MCC of 0.153, whereas SVM obtained an AUC of 0.497, F1-Score of 0.743, Precision of 0.695, and MCC of 0.000. The near-zero MCC and low AUC of SVM suggest that the model struggled to distinguish minority classes under extreme class imbalance despite achieving high accuracy. These findings highlight the importance of employing multiple evaluation metrics beyond accuracy when assessing classification performance on imbalanced datasets. Furthermore, the results indicate that Multinomial Naïve Bayes provided better performance than SVM under the dataset characteristics and experimental configuration used in this study.]]>
