This Author published in this journals
All Journal semanTIK
Imamah Mailah
Universitas Islam Madura

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Analisis Sentimen Publik Debat Pilkada Pamekasan menggunakan BERT Imamah Mailah; Moh. Aminollah Hamzah; Hozairi
SemanTIK : Teknik Informasi Vol. 11 No. 2 (2025): SemanTIK : Teknik Informasi
Publisher : Informatics Engineering Department of Halu Oleo University

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.55679/semantik.v11i2.152

Abstract

Pemilihan kepala daerah merupakan momen penting dalam demokrasi yang memunculkan beragam opini publik di media sosial. Debat calon bupati dan wakil bupati Pamekasan tahun 2024 menjadi perhatian masyarakat dan menghasilkan banyak komentar daring. Penelitian ini bertujuan untuk menganalisis sentimen publik terhadap debat tersebut menggunakan pendekatan deep learning berbasis transformer. Data penelitian berupa 818 komentar dari YouTube dan TikTok yang diperoleh melalui web scraping. Tahapan penelitian meliputi pembersihan data, case folding, tokenisasi, serta terjemahan. Proses pelabelan sentimen dilakukan dengan TextBlob, sedangkan klasifikasi menggunakan model DistilBERT yang telah di-fine-tune. Hasil penelitian menunjukkan model mampu mengklasifikasikan komentar menjadi tiga kategori, yaitu positif, netral, dan negatif, dengan akurasi 80% serta F1-score tertinggi 0,91 pada kelas positif. Sebagian besar komentar tergolong netral (44,03%), diikuti positif (37,03%) dan negatif (18,96%). Temuan ini menunjukkan bahwa respon publik cenderung biasa tanpa ekspresi emosional yang kuat. Penelitian ini menyimpulkan bahwa model berbasis transformer efektif untuk menganalisis opini publik dalam konteks politik lokal, sehingga dapat membantu pengambil kebijakan, pengamat politik, maupun tim kampanye memahami persepsi masyarakat secara lebih cepat dan akurat. Regional elections are a crucial moment in democracy that generate diverse public opinions on social media. The 2024 Pamekasan regent and deputy regent candidate debate attracted public attention and sparked many online comments. This study aims to analyze public sentiment toward the debate using a transformer-based deep learning approach. The dataset consists of 818 comments collected from YouTube and TikTok through web scraping. The research process included data cleaning, case folding, tokenization, and translation. Sentiment labeling was carried out using TextBlob, while classification employed a fine-tuned DistilBERT model. The results show that the model successfully categorized comments into three sentiment classes—positive, neutral, and negative—with an accuracy of 80% and the highest F1-score of 0.91 in the positive class. Most comments were classified as neutral (44.03%), followed by positive (37.03%) and negative (18.96%). These findings indicate that the majority of the public responded in a neutral manner without strong emotional bias. This study concludes that transformer-based models are effective in analyzing public opinion in local political contexts, providing valuable insights for policymakers, political observers, and campaign teams to better understand community perceptions quickly and accurately.