Validation of the Use of a Large Language Model for Detecting Sentiment in Student Course Evaluation

The authors aimed to validate the use of a bidirectional encoder representations from transformers (BERT) model for detecting sentiment in medical student course evaluations, addressing concerns about bias and errors in artificial intelligence tools. By comparing the sentiment analysis results of the BERT model with human coders across multiple institutions, they found that the model's interrater reliability was comparable to that of human evaluators. This study supports the potential application of NLP methods in health professions education for analyzing student feedback.