Analisis Sentimen Debat Capres dan Cawapres Indonesia 2024 Menggunakan Metode Bert

Abstract

The 2024 Indonesian Presidential and Vice Presidential Candidate Debate, broadcast on the YouTube channel of the Indonesian General Elections Commission (KPU), generated thousands of public comments reflecting societal sentiment toward the candidates, yet these opinions have not been systematically examined. This study aims to analyze the sentiment of YouTube comments on the debate using the Bidirectional Encoder Representations from Transformers (BERT) method. A total of 2,500 comments were collected via the YouTube Data API from five debate sessions, then cleaned through a pre-processing stage and automatically labeled using the InSetLexicon dictionary into three categories: positive, negative, and neutral. After removing duplicate data, 2,362 clean records were obtained and divided into training and testing data with an 80:20 ratio. The IndoBERT model was then fine-tuned to classify comment sentiment. The results show that the sentiment distribution was dominated by positive comments (1,107), followed by negative (660) and neutral (595) comments. Model evaluation using a confusion matrix yielded an accuracy of 76.5%, with the highest precision, recall, and F1-score values in the positive class at 87.78%, 85.39%, and 86.57%, respectively. These findings indicate that the BERT method is fairly effective in analyzing the sentiment of Indonesian-language comments in the context of political debates on YouTube, while also contributing a mapping of public opinion toward candidates and debate issues that can serve as a reference for future Indonesian-language political sentiment analysis research.