Sentiment Analysis of Students Reviews on Campus Curriculum for the Workplace Using Bidirectional Encoder Representations from Transformers BERT Algorithm

Abstract

Higher education curricula play a crucial role in preparing graduates for the workforce; however, a gap often exists between theoretical instruction and the practical demands of industry. This study evaluates curriculum effectiveness through sentiment analysis of students reviews using the Bidirectional Encoder Representations from Transformers (BERT) model. A total of 250 students reviews were collected via open-ended questionnaires and processed using Python with libraries such as NLTK, Transformers, and Pandas for text cleaning, normalization, and tokenization. Twenty percent of the data were manually labeled, achieving a Cohen’s Kappa coefficient of 0.8405, indicating excellent inter-annotator agreement. The IndoBERT model—implemented using the Torch and Scikit-learn libraries—was trained to classify sentiments into positive, negative, and neutral categories. Results show that 82.7% of reviews were positive, praising the curriculum’s relevance to job readiness and technical skills; 12.7% were negative, highlighting insufficient practical content and soft skills development; and 4.5% were neutral. This research demonstrates that BERT-based sentiment analysis using Python is effective for curriculum evaluation, providing data-driven insights to help academic institutions enhance curriculum alignment with industry needs