Penelitian ini mengkaji respon masyarakat terhadap kebijakan kenaikan Pajak Pertambahan Nilai (PPN) dari 11% menjadi 12%. Kebijakan tersebut menimbulkan beragam pendapat di kalangan warganet, khususnya pada platform media sosial X (Twitter). Tujuan penelitian ini adalah membangun dataset dan model algoritma Support Vector Machine (SVM) dengan kernel sigmoid, serta mengevaluasi tingkat akurasinya dalam menganalisis sentimen masyarakat terkait kebijakan kenaikan PPN 12%. Data dikumpulkan melalui teknik scraping pada periode Oktober 2024 hingga April 2025 dan menghasilkan 1.051 data. Proses preprocessing meliputi case folding, tokenisasi, penghapusan stopwords, dan stemming. Selanjutnya, data diberi label sentimen dan dilakukan pembobotan kata menggunakan metode Term Frequency–Inverse Document Frequency (TF-IDF). Model SVM kemudian dilatih dan dievaluasi menggunakan teknik K-Fold Cross Validation, dengan hasil confusion matrix menunjukkan tingkat akurasi sebesar 91%. Selain itu, nilai F1-score diperoleh sebesar 89% untuk sentimen negatif, 92% untuk sentimen netral, dan 95% untuk sentimen positif. Hasil penelitian ini membuktikan bahwa penerapan machine learning, khususnya SVM dengan kernel sigmoid, dapat secara efektif menangkap persepsi publik melalui media sosial, sehingga dapat dijadikan tolok ukur bagi pemerintah dalam merumuskan dan mengevaluasi kebijakan.

This study examines public responses to the policy of increasing the Value Added Tax (VAT) rate from 11% to 12%. The policy sparked diverse opinions among netizens, particularly on the social media platform X (Twitter). The objective of this research is to build a dataset and develop a Support Vector Machine (SVM) model using a sigmoid kernel, as well as to evaluate its accuracy in analyzing public sentiment toward the VAT increase. Data were collected through scraping techniques during the period from October 2024 to April 2025, resulting in 1,051 entries. The preprocessing stage involved case folding, tokenization, stopword removal, and stemming. The data were then labeled based on sentiment categories and weighted using the Term Frequency–Inverse Document Frequency (TF-IDF) method. The SVM model was trained and evaluated using K-Fold Cross Validation, with the confusion matrix showing an accuracy of 91%. Furthermore, the F1-scores achieved were 89% for negative sentiment, 92% for neutral sentiment, and 95% for positive sentiment. These findings demonstrate that machine learning techniques, particularly SVM with a sigmoid kernel, can effectively capture public perceptions through social media, providing valuable insights for the government in formulating and evaluating policies..