<![CDATA[Abstrak - M-Pajak adalah aplikasi digital yang dikembangkan oleh Direktorat Jenderal Pajak (DJP) untuk membantu masyarakat dalam mengelola perpajakan. Namun, aplikasi ini menerima banyak ulasan negatif di Google PlayStore. Penelitian ini bertujuan untuk menganalisis sentimen dari ulasan tersebut, dengan membagi ulasan menjadi sentimen positif dan negatif. Tiga algoritma klasifikasi yang digunakan adalah Support Vector Machine (SVM), Naive Bayes, dan K-Nearest Neighbors (KNN). Evaluasi dilakukan menggunakan teknik 10-Fold Cross Validation untuk mengukur accuracy, recall, dan precision. Hasilnya menunjukkan bahwa SVM memiliki kinerja terbaik dengan accuracy 86,41%, recall positif 42,20%, recall negatif 98,12%, precision positif 85,61%, dan precision negatif 86,50%. Naive Bayes berada di posisi kedua dengan accuracy 66,95%, recall positif 71,81%, recall negatif 65,66%, precision positif 35,65%, dan precision negatif 89,79%. Sementara itu, KNN memiliki performa terendah dengan accuracy 45,01%, recall positif 75,35%, recall negatif 36,97%, precision positif 24,05%, dan precision negatif 84,99%. Penelitian ini menyimpulkan bahwa SVM merupakan metode yang paling efektif untuk analisis sentimen ulasan M-Pajak.Kata kunci: Support Vector Machine, Naive Bayes, KNN, Sentimen. Abstract - M-Pajak is a digital application developed by the Direktorat Jenderal Pajak (DJP) to assist the public in managing their taxes. However, the app has received numerous negative reviews on Google PlayStore. This study aims to analyze these reviews by classifying them into positive and negative sentiments. Three classification algorithms were used: Support Vector Machine (SVM), Naive Bayes, and K-Nearest Neighbors (KNN). The evaluation was conducted using 10-Fold Cross Validation to measure accuracy, recall, and precision. The results showed that SVM performed the best, achieving an accuracy of 86.41%, with a positive recall of 42.20%, a negative recall of 98.12%, a positive precision of 85.61%, and a negative precision of 86.50%. Naive Bayes ranked second with 66.95% accuracy, 71.81% positive recall, 65.66% negative recall, 35.65% positive precision, and 89.79% negative precision. KNN had the lowest performance, with 45.01% accuracy, 75.35% positive recall, 36.97% negative recall, 24.05% positive precision, and 84.99% negative precision. The study concludes that SVM is the most effective method for sentiment analysis of M-Pajak reviews.Keywords: Support Vector Machine, Naive Bayes, KNN, Sentiment.]]>
