<![CDATA[The 2024 Indonesian presidential election inaugurated the Prabowo Subianto–Gibran Rakabuming Raka administration, whose early performance has been widely discussed on digital social networks, particularly X (Twitter). This study evaluates public sentiment toward the administration's performance up to June 30, 2025 using an IndoBERT-based text classification approach. A total of 2,612 public posts were collected via web scraping and processed through text preprocessing steps (noise removal, slang correction, normalization, and lemmatization). The data were labeled into three sentiment classes (positive, neutral, and negative) and split into training, validation, and test sets (2,092 / 418 / 105). The fine-tuned IndoBERT model achieved an overall test accuracy of 0.78, with the highest F1-score on the negative class (0.82), followed by neutral (0.76) and positive (0.75). The confusion matrix indicates that neutral posts are more frequently confused with positive posts, suggesting that neutral sentiment remains harder to separate in politically nuanced and noisy social-media text. This study also compares IndoBERT with a traditional baseline (TF-IDF + SVM using polynomial kernel). Results show that IndoBERT (78% accuracy) significantly outperforms SVM (72.19%), particularly in detecting negative sentiment (F1: 0.82 vs 0.72), demonstrating superior contextual understanding of politically nuanced text. This work also highlights methodological and ethical considerations for political opinion mining, including representativeness limits of X users and privacy-preserving handling of public posts. Future work should expand the dataset, address class imbalance, and explore additional transformer-based architectures to strengthen generalizability and benchmarking.]]>
