Cutting-edge Technologies for Analyzing Student Feedback to Inform Institutional Decision-making in Higher Education

Abstract

Aspect-Based Sentiment Analysis (ABSA) has emerged as a powerful tool for deriving actionable insights from qualitative feedback in education. This study presents a multitask learning framework to analyze student evaluations of teaching (SET) by extracting and classifying opinions on specific aspects of teaching performance. Leveraging a novel and first open-sourced dataset of 6,025 Spanish-language comments, the proposed framework integrates opinion segmentation and multi-label classification to capture nuanced feedback on nine predefined aspects, such as "Teaching Quality" and "Classroom Atmosphere." Applications of this approach extend beyond SET analysis, offering valuable insights for course improvement, faculty assessment, and institutional decision-making in higher education. The paper compares the performance of fine-tuned transformers (BERT and RoBERTa) with large language models (LLMs), including GPT-4o, GPT4o-mini, and LLama-3.1-8B, using both fine-tuned and Few-shot Chain of Thought (CoT) methodologies. Evaluation results reveal that fine-tuned GPT-4o outperformed all other models, achieving a weighted F1-score of 0.69 for positive aspects and 0.79 for negative aspects, while Few-shot CoT approaches demonstrated competitive performance with greater scalability and interpretability. Our findings demonstrate the framework's potential to transform unstructured feedback into structured insights, aiding educators and institutions in enhancing teaching quality and student engagement.