The project investigates which NLP models, including large language models and transformer-based classifiers, perform best when evaluating student course evaluation comments. Previous attempts produced unreliable reports that often assigned generic comments to multiple incorrect categories and lacked a clearly defined rubric for scoring and interpretation. To address this, the project introduces a structured topic framework for course feedback and evaluates how well different models can classify comments into meaningful instructional categories such as course organization, workload, clarity, assessments, communication, and inclusivity.

By comparing models such as Gemma, Llama 3, DistilRoBERTa, and RoBERTa-based approaches, the project aims to identify which methods are most effective for multi-label topic classification of open-ended educational feedback. The broader goal is to improve the consistency, interpretability, and usefulness of automated course evaluation analysis so that instructors and institutions can better understand student experiences at scale.