<![CDATA[This study aims to analyze the sentiment of Indonesian-language YouTube comments related to artificial intelligence (AI) in the field of education using a Transformer-based deep learning approach, namely Bidirectional Encoder Representations from Transformers (BERT) and IndoRoBERTa. The research data were obtained through the YouTube Data API, consisting of 10,834 comments reflecting public opinion on the implementation of AI in education. The dataset was manually labeled into three sentiment categories: positive, neutral, and negative, followed by a preprocessing stage including case folding, text cleaning, text normalization, tokenization, and topic filtering. The experimental results show that in the baseline scenario without fine-tuning, both models achieved low performance with accuracy below 41%. However, after fine-tuning, a significant improvement was observed, where IndoRoBERTa achieved an accuracy of 91.54% with an F1-score of 0.9134, while BERT reached an accuracy of 84.63% with an F1-score of 0.8413. These results indicate that Transformer-based models adapted to specific datasets are capable of better capturing the contextual and linguistic characteristics of informal and unstructured Indonesian text. In addition, IndoRoBERTa demonstrates more stable performance in handling class imbalance compared to BERT. Overall, this study demonstrates that Transformer-based approaches are effective for sentiment analysis in social media and can be used to more accurately and comprehensively understand public perceptions of the implementation of artificial intelligence in education.]]>
Copyrights © 2026
