Scholarlink

In the context of intelligent tutoring systems, leveraging structured academic records for personalization holds substantial promise. This study introduces TranscriptQA, a novel dataset composed of academic transcript data and aligned educational queries designed to evaluate the capacity of large language models (LLMs) to generate contextually adapted responses. A Retrieval-Augmented Generation (RAG) framework is proposed, dynamically integrating transcript-relevant content into the LLM generation process. Unlike prior methods, this approach enables scalable and adaptive personalization based on formal academic performance records. Experimental evaluations, using BERTScore, BLEURT, and METEOR metrics, demonstrate that transcript-informed LLMs significantly outperform generic baselines in both semantic relevance and response appropriateness, showing notable gains of +0.028 in BERTScore, +0.159 in BLEURT, and +0.160 in METEOR over non-personalized models. The findings establish a foundation for integrating structured learner data into generative AI systems, supporting personalized learning at scale.

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

Adaptive LLM Responses with RAG: A Dataset and Methodology for Personalized Education via Academic Transcripts

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vinh Dinh Nguyen,
  • Nhan Huu Tran,
  • Phong Van Nguyen,
  • Khoa Anh Dao

摘要

In the context of intelligent tutoring systems, leveraging structured academic records for personalization holds substantial promise. This study introduces TranscriptQA, a novel dataset composed of academic transcript data and aligned educational queries designed to evaluate the capacity of large language models (LLMs) to generate contextually adapted responses. A Retrieval-Augmented Generation (RAG) framework is proposed, dynamically integrating transcript-relevant content into the LLM generation process. Unlike prior methods, this approach enables scalable and adaptive personalization based on formal academic performance records. Experimental evaluations, using BERTScore, BLEURT, and METEOR metrics, demonstrate that transcript-informed LLMs significantly outperform generic baselines in both semantic relevance and response appropriateness, showing notable gains of +0.028 in BERTScore, +0.159 in BLEURT, and +0.160 in METEOR over non-personalized models. The findings establish a foundation for integrating structured learner data into generative AI systems, supporting personalized learning at scale.