<![CDATA[Digital transformation of public services has driven the Indonesian National Police to develop the Polri Super App, yet faces user acceptance challenges reflected in diverse reviews. This study aims to compare the performance of Support Vector Machine (SVM) and Naïve Bayes algorithms in classifying user sentiment of the Polri Super App. The research utilized 3,997 reviews from Apple Store undergoing comprehensive preprocessing including normalization, tokenization, stopword removal, and Sastrawi stemming. Sentiment labeling employed InSet Lexicon, yielding 55.0% positive and 45.0% negative reviews. Feature extraction used TF-IDF method with 80:20 data split for training and testing. Evaluation results demonstrate SVM significantly outperforms Naïve Bayes with 91.5% versus 79.0% accuracy (12.5 percentage points difference). SVM maintains balanced F1-scores of 90.6% (negative) and 92.2% (positive), while Naïve Bayes exhibits imbalance with 87.2% recall (negative) but only 72.3% (positive). SVM's superiority stems from hyperplane optimization capability in handling high-dimensional text data without rigid feature independence assumptions. The study recommends SVM implementation for police digital service sentiment monitoring systems and exploration of ensemble algorithms and deep learning for future research.]]>
Copyrights © 2026
