<![CDATA[User reviews of the Gojek application on the Google Play Store contain various types of information regarding problems experienced by users. However, most previous studies have focused on sentiment analysis and have not been able to identify specific problem types. This study aims to classify problem types in Gojek user reviews into five categories: login, transaction, system disruption, feature error, and other issues. Data were collected through web scraping from the Google Play Store between January 2024 and April 2026, targeting 5,000 reviews. After removing empty and duplicate records, 3,753 reviews were retained for analysis. Labeling was performed using a keyword-based rule-based approach, followed by preprocessing stages including case folding, cleaning, tokenization, stopword removal, and stemming. Feature representation was conducted using TF-IDF with a maximum of 2,500 features. Class imbalance in the training data was addressed using Random Over Sampling (ROS), while the dataset was split using an 80:20 ratio through stratified sampling. This study compares the performance of Support Vector Machine (SVM) and Naïve Bayes classifiers using accuracy, precision, recall, F1-score, and ROC-AUC metrics, with model validation performed through 5-fold Stratified K-Fold Cross Validation. The results show that SVM achieved the best performance, with an accuracy of 0.846, precision of 0.872, recall of 0.846, F1-score of 0.856, ROC-AUC of 0.903, and an average cross-validation F1-score of 0.8509. In contrast, Naïve Bayes achieved an accuracy of 0.555, precision of 0.828, recall of 0.555, F1-score of 0.635, ROC-AUC of 0.836, and an average cross-validation F1-score of 0.6281. These results indicate that SVM performs better in classifying problem types in Gojek user reviews.]]>
