<![CDATA[The rapid growth of e-commerce in Indonesia, particularly within Shopee’s fashion category, has generated a large volume of customer reviews that can serve as valuable sources of business insights. However, sentiment classification of these reviews is challenged by the extensive use of informal slang expressions and imbalanced sentiment distributions. This study develops an Aspect-Based Sentiment Analysis (ABSA) model by comparing the performance of Multinomial Naïve Bayes (MNB) and Support Vector Machine (SVM), while integrating slang normalization and the Synthetic Minority Oversampling Technique (SMOTE). A dataset of 5,877 customer reviews was analyzed using an 80:20 train–test split and evaluated through a confusion matrix with accuracy, precision, recall, and F1-score metrics. The results indicate that, without preprocessing, SVM achieved an accuracy of 0.852, outperforming MNB with an accuracy of 0.722. After applying slang normalization and SMOTE, the performance of both models improved substantially. MNB achieved an accuracy of 0.853, while SVM attained the highest performance with an accuracy of 0.938, precision of 0.939, recall of 0.938, and F1-score of 0.937. These findings demonstrate that integrating slang normalization and SMOTE effectively enhances aspect-based sentiment classification, with SVM providing the best performance on Shopee fashion product reviews.]]>
Copyrights © 2026
