Analyzing Student Sentiments and Insights on Generative AI for Independent Learning in Universities

Abstract

Transformations in higher education brought about by Generative AI have significantly changed how university students’ access, comprehend, and develop learning materials. This study explores Indonesian university students’ perceptions and experiences regarding the use of Generative AI for independent learning, employing qualitative surveys together with sentiment analysis powered by machine learning. Data were collected from open-ended questionnaires and analyzed using four key algorithms, such as Naive Bayes, Logistic Regression, Random Forest, and Linear SVM, to classify student sentiments towards generative AI technologies. These four classical machine learning models were employed as baseline algorithms commonly used in sentiment analysis to benchmark performance on small, imbalanced educational datasets before applying more complex transformer-based methods. In addition to quantitative analysis, this study also implements thematic analysis of open-ended responses to identify prominent issues, challenges, and student recommendations concerning the use of generative AI in learning. Evaluation results identified Linear SVM as the most consistent model, with the highest weighted F1-score (0.63), although all models showed limitations in detecting negative sentiment due to class imbalance (only three negative samples out of forty responses), which affected model generalization. Key findings indicate that students perceive Generative AI as a supportive tool that accelerates understanding, creativity, and reference searching; however, they remain wary of risks related to dependency, reduced originality, and academic integrity dilemmas. This article recommends the implementation of ethical policy, AI digital literacy training, and enhancement of campus infrastructure to ensure that AI technologies enrich the learning process without compromising student independence and integrity.