<![CDATA[Digital transformation in higher education requires continuous, data-driven evaluation of student satisfaction to ensure service quality and institutional improvement. Student feedback, commonly expressed in free-text form, represents a rich source of information that reflects students’ perceptions of academic services. However, the unstructured nature of textual feedback poses challenges for large-scale analysis. This study aims to analyze student feedback sentiment at Ary Ginanjar University using a Natural Language Processing (NLP) approach based on the IndoBERT model. The research methodology follows the CRISP-DM framework, encompassing business understanding, data understanding, data preparation, modeling, evaluation, and deployment. During the data preparation stage, a weak labeling strategy is employed to automatically assign initial sentiment labels using a pre-trained IndoBERT model, followed by manual correction to improve label quality. This human-in-the-loop approach enables the construction of a reliable labeled dataset while reducing annotation effort. The dataset is classified into three sentiment categories: positive, neutral, and negative. Text preprocessing includes normalization, punctuation removal, and tokenization using the IndoBERT tokenizer. Subsequently, IndoBERT is fine-tuned for sentiment classification, and model performance is evaluated using accuracy, precision, recall, and F1-score metrics. To gain deeper insights into dissatisfaction factors, topic modeling using Latent Dirichlet Allocation (LDA) is applied to feedback classified as negative, resulting in five dominant topics. These topics reveal key issues related to teaching structure, learning effectiveness, workload, and classroom management. The findings provide actionable strategic recommendations to support data-driven decision-making and continuous improvement at Ary Ginanjar University.  ]]>
