<![CDATA[The rapid growth of e-commerce in Indonesia has led to an increase in user interactions in the form of reviews and opinions on services and products. These textual data contain valuable information that can be processed through sentiment analysis to better understand user perceptions. This study aims to compare the effectiveness of Term Frequency–Inverse Document Frequency (TF-IDF) and Bag of Words (BoW) feature extraction methods in classifying user sentiments, as well as to evaluate the performance of Support Vector Machine (SVM) and Random Forest (RF) algorithms on Shopee and Tokopedia platforms. A total of 5,000 user reviews were analyzed through text preprocessing, lexicon-based sentiment labeling, application of TF-IDF and BoW feature extraction methods, model training using SVM and RF algorithms, and performance evaluation using accuracy, precision, recall, and F1-score metrics. The experimental results show the combination of BoW and SVM achieved the highest accuracy of 90% on Shopee reviews, making it the most optimal configuration in this study. Additionally, in Tokopedia reviews, the same configuration (BoW and SVM) also produced a strong accuracy of 88%. In general, the SVM algorithm showed more stable performance than RF, while the BoW method proved to be more effective (measured at up to 90% accuracy) in representing this Indonesian-language e-commerce review data. These findings contribute to the development of more accurate sentiment analysis systems in the local e-commerce domain.]]>
