<![CDATA[Abstrak - Perkembangan platform digital seperti YouTube telah menciptakan ruang publik baru untuk diskusi politik, dimana kolom komentar menjadi cerminan langsung tanggapan masyarakat terhadap konten politik. Penelitian ini menganalisis sentimen publik terhadap video Presiden Prabowo Menjawab di YouTube menggunakan algoritma Multinomial Naïve Bayes yang dioptimasi dengan seleksi fitur Chi-Square dan teknik SMOTE untuk penanganan ketidakseimbangan data. Data diperoleh melalui web scraping sebanyak 4.649 komentar dari tiga channel YouTube terkemuka. Preprocessing teks dilakukan melalui cleaning, case folding, normalisasi, tokenizing, stopword removal, dan stemming. Hasil penelitian menunjukkan bahwa model mencapai akurasi sebesar 67% dengan precision kelas positif 72% dan recall kelas negatif 81%. Hasil analisis mengungkapkan bahwa model lebih efektif dalam mengidentifikasi komentar negatif dibandingkan positif, yang merefleksikan karakteristik unik pola linguistik dalam diskusi politik di platform digital. Penelitian ini memberikan kontribusi dalam pengembangan sistem analisis sentimen otomatis untuk konten politik serta menjadi dasar untuk pemantauan opini publik di era digital.Kata kunci: Analisis Sentimen; Naïve Bayes; Chi-Square; SMOTE; Abstract - The development of digital platforms such as YouTube has created a new public space for political discussion, where comment sections directly reflect public responses to political content. This study analyzes public sentiment toward the video “Presiden Prabowo Menjawab” (President Prabowo Responds) on YouTube using a Multinomial Naïve Bayes algorithm optimized with Chi-Square feature selection and SMOTE technique for data imbalance handling. Data was obtained through web scraping of 4,649 comments from three leading YouTube channels. Text preprocessing was carried out through cleaning, case folding, normalization, tokenizing, stopword removal, and stemming. The results show that the model achieved an accuracy of 67% with a positive class precision of 72% and a negative class recall of 81%. The analysis results reveal that the model is more effective in identifying negative comments than positive ones, reflecting the unique characteristics of linguistic patterns in political discussions on digital platforms. This research contributes to the development of an automatic sentiment analysis system for political content and serves as a basis for monitoring public opinion in the digital era.Keywords: Sentiment Analysis; Naïve Bayes; Chi-Square; SMOTE;]]>
