<![CDATA[The rapid growth of mobile applications necessitates a better understanding of user feedback via app store reviews. This study analyzes 1,000 Indonesian-language Google Play Store reviews of the Instagram application using the Support Vector Machine (SVM) algorithm to classify sentiments into positive, negative, and neutral categories. The primary objectives are to evaluate the effectiveness of SVM and investigate the impact of neutral sentiments on model performance. The methodology involved preprocessing steps such as text cleaning, slang normalization, stop-word removal, and stemming with Sastrawi, followed by TF-IDF feature extraction. A LinearSVC model was optimized using GridSearchCV and five-fold cross-validation. The model achieved 64.5% accuracy and a Macro-F1 score of 0.4701, outperforming Multinomial Naïve Bayes and Logistic Regression baselines. Further analysis revealed that neutral sentiments significantly affect performance; removing this class increased Macro-F1 to 0.6965. Additionally, probability thresholding and class-weight balancing improved neutral-class recognition, raising its F1-score from 0.1429 to 0.1778. These findings indicate that while SVM is effective for Indonesian app-review sentiment analysis, neutral sentiment remains a classification challenge requiring specific handling strategies.]]>
Copyrights © 2026
