Scholarlink

Learner performance estimation is a critical topic in education, offering the potential to model learner understanding. Large language models (LLMs) have recently gained attention for student modeling and performance estimation. However, because LLMs are pre-trained on vast amounts of information, they often possess knowledge that exceeds that of individual learners. As a result, an LLM may correctly answer questions without relying on the content explicitly provided by the learner.We propose a learner response estimation method based on the learner’s note-driven Retrieval Augmented Generation (RAG) to discuss how we can model LLMs as student models. In an on-demand classroom environment, we experimentally collect participants’ notes as learners acquire knowledge. We evaluate whether notes are helpful in LLM-based learner performance estimation and investigate the necessary adjustments for LLMs.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Note-Driven RAG for Learner Performance Estimation via Controlling LLM Knowledge

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tsubasa Minematsu,
  • Atsushi Shimada

摘要

Learner performance estimation is a critical topic in education, offering the potential to model learner understanding. Large language models (LLMs) have recently gained attention for student modeling and performance estimation. However, because LLMs are pre-trained on vast amounts of information, they often possess knowledge that exceeds that of individual learners. As a result, an LLM may correctly answer questions without relying on the content explicitly provided by the learner.We propose a learner response estimation method based on the learner’s note-driven Retrieval Augmented Generation (RAG) to discuss how we can model LLMs as student models. In an on-demand classroom environment, we experimentally collect participants’ notes as learners acquire knowledge. We evaluate whether notes are helpful in LLM-based learner performance estimation and investigate the necessary adjustments for LLMs.