<![CDATA[The rapid growth of e-commerce generates a large volume of Shopee customer reviews, making manual evaluation inefficient and inconsistent. This study addresses the need for automated text classification to identify review polarity and support seller service evaluation. A labeled dataset of 800 Shopee reviews (600 positive, 200 negative) was collected from Kaggle. The texts were preprocessed using case folding, tokenization, stopword removal, and stemming to reduce noise and standardize writing. Reviews were converted into TF-IDF feature vectors (max_features=5,000; n-gram range 1-2) and classified using a Support Vector Machine (SVM) with an RBF kernel and class-weight balancing. The data were split with stratified sampling (80:20) to preserve label proportions in training and testing sets. Model performance was measured using accuracy, precision, recall, F1-score, and a confusion matrix. The classifier achieved 83.75% accuracy and produced stronger results for positive reviews, while negative reviews were more frequently misclassified due to class imbalance and diverse complaint expressions. As supporting documentation, a subset of reviews was prepared as screenshots and annotated in Roboflow, while the main sentiment evaluation remained based on TF-IDF feature extraction and SVM classification. Overall, TF-IDF combined with SVM provides a practical approach for monitoring customer sentiment and prioritizing seller service improvements over time.]]>
Copyrights © 2026
