Evaluating Cognitive Performance Through Prompt-Based Methods Using LLM in Education

Abstract

Bloom's taxonomy is widely used to characterize student cognitive performance, but manual coding does not scale to AI-tutored dialogue. We evaluate prompt-engineering strategies that let an LLM automatically classify student responses across Bloom levels, finding both that the task is feasible and that prompt design has substantial effect on accuracy.

Problem & Motivation

Bloom's taxonomy is a widely used framework for characterizing the cognitive level of student responses, but manual coding (remember / understand / apply / analyze / evaluate / create) does not scale. As AI-mediated learning conversations proliferate, automatic and reliable assessment of student cognitive performance becomes a central methodological problem.

Method

We explored prompt-engineering strategies that allow a large language model to automatically evaluate cognitive performance in educational contexts. Carefully designed prompts guide the LLM to classify each student response against Bloom's taxonomy.

Findings

• Prompt-engineered LLMs are effective at evaluating student cognitive level.
• Prompt design choices materially affect classification accuracy.
• Automated cognitive assessment offers a tractable path for large-scale learning analytics.

Implications

Automated cognitive-level assessment lets instructors see beyond raw correctness and look at how students think. This is particularly relevant for cultivating higher-order thinking, where instruction must respond not only to wrong answers but to shallow patterns of reasoning.

Citation

@inproceedings{furqon2025cognitive_eval,
  author    = {Elvin Nur Furqon and Chia-Kai Chang},
  title     = {Evaluating Cognitive Performance Through Prompt-Based Methods Using {LLM} in Education},
  booktitle = {Proc. IEEE Int. Conf. on Advanced Learning Technologies (ICALT)},
  pages     = {209--211},
  year      = {2025},
  month     = jul,
  doi       = {10.1109/ICALT64023.2025.00065},
}