A Two-Phase LLM-Based Procedure for Processing Spanish-Language Qualitative Teaching Feedback in Higher Education

Bravo E.R. · EDUNINE 2026 - 10th IEEE World Engineering Education Conference: The Role of Engineering and Computer Technology as Artificial Intelligence turns into Reality: A Challenge for Professional Education, Proceedings (2026) · DOI: 10.1109/edunine62390.2026.11547053

Resumen

Purpose: This study addresses the challenge of efficiently processing Spanish-language qualitative teaching feedback in higher education by developing a systematic approach combining Large Language Models (LLMs) with human oversight. Method: A two-phase procedure was implemented: Phase 1 utilized ChatGPT-4 for generating feedback categories from student comments, followed by expert validation. Phase 2 employed the validated framework to classify individual comments through LLM processing and human refinement. Findings: The procedure yielded eighteen distinct categories encompassing pedagogical and interpersonal dimensions of teaching. The implementation effectively categorizes diverse student feedback while maintaining analytical rigor through human oversight. Originality/Value: This study presents an innovative, resource-efficient approach for Spanish-language institutions to process qualitative feedback, addressing limitations in current NLP solutions while ensuring pedagogical relevance. Research limitations: Current limitations of the procedure include prompt dependency, execution variability, context window restrictions, and the black box nature of LLMs, which impede complete understanding and auditability of the analytical process. Practical implications: The framework enables systematic analysis of qualitative feedback for evidence-based decision-making in faculty development and course improvement initiatives.

Instituciones: Universidad del Pacífico