Analisis Sentimen Publik Terhadap Magang Berdampak 2025 Di Platform X/Twitter Menggunakan Model Indobert

Abstract

Social media serves as a public discourse platform where individuals express views on various policies, including higher education. The Impactful Internship Program 2025, as a continuation of the Merdeka Belajar Kampus Merdeka policy, has generated diverse responses on the X platform, making sentiment analysis necessary to systematically capture public perceptions. This study analyzes public sentiment toward the Impactful Internship Program 2025 using the IndoBERT model and evaluates the impact of hyperparameter configurations on classification performance. The dataset consists of Indonesian-language posts from the X platform classified into positive, negative, and neutral sentiments. Results show that the best model was obtained at the 6th epoch using a learning rate of 2e-5 and a batch size of 32, achieving an accuracy of 0.7481, a precision of 0.7720, a recall of 0.7481, and an f1-score of 0.7562. The analysis indicates that neutral sentiment dominates public discourse, followed by negative and positive sentiments, suggesting informative discussions accompanied by criticism of the program’s implementation. These findings highlight the importance of hyperparameter selection and the challenges of sentiment analysis on imbalanced and context-dependent social media data.