Opinion Mining of Students Feedback on Online Education Using BiLSTM–Attention and Transformer Models

Online education generates large amounts of student feedback that requires systematic analysis. This study develops an opinion mining system to classify sentiments in student reviews about online learning platforms. We compare BiLSTM models with attention mechanisms against transformer models like BERT and IndoBERT. Experiments use public datasets containing thousands of labeled student comments from university surveys and MOOC platforms. Preprocessing includes tokenization, stopword removal, and class balancing. Results show transformer models achieve up to 93% accuracy and superior F1scores compared to traditional baselines like SVM. IndoBERT performs best for multilingual feedback common in Indian contexts. Attention visualizations highlight key issues such as technical problems and lack of interaction. Findings offer educators actionable insights to improve course design and platform usability. The framework supports real-time monitoring of student opinions in virtual learning environments.