Scholarlink

The growing diversity of professional training resources and learning objects, from MOOCs and micro-credentials to certified vocational programs, poses significant challenges for learners seeking pedagogically coherent and personalized pathways. In this paper, we introduce CLARE (Category-aware Learning object recommendation with Augmented REtrieval), a hybrid framework that unites sequential category prediction with retrieval-augmented generation (RAG) via large language models (LLMs) to guide learners toward relevant, pedagogically sound opportunities. At its core, CLARE leverages structured learning categories as semantic anchors to align user intent, educational taxonomies, and content retrieval. The framework comprises two key modules: (1) a BERT4Rec-based model that forecasts the next relevant learning categories from a learner’s historical interactions and (2) a category-aware prompt engineering strategy enhanced by In-Context Learning (ICL) that supplies the LLM with representative learning patterns to generate highly personalized learning recommendations. We evaluate CLARE on a real-world training dataset under both full-context and cold-start conditions. Our results demonstrate that integrating category-driven retrieval and ICL-augmented prompt design substantially improves recommendation relevance and coverage. These findings underscore the promise of combining LLM-powered RAG pipelines with domain-structured metadata to deliver adaptive, scalable, and pedagogically coherent learning guidance.

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

CLARE: A Category-Aware RAG-Based Framework for Recommending Learning Objects in Education and Professional Training

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ichrak Ennaceur,
  • Haytham Elghazel,
  • Alexandre Aussem,
  • Guillaume Lefebvre,
  • Matthieu Sonnati

摘要

The growing diversity of professional training resources and learning objects, from MOOCs and micro-credentials to certified vocational programs, poses significant challenges for learners seeking pedagogically coherent and personalized pathways. In this paper, we introduce CLARE (Category-aware Learning object recommendation with Augmented REtrieval), a hybrid framework that unites sequential category prediction with retrieval-augmented generation (RAG) via large language models (LLMs) to guide learners toward relevant, pedagogically sound opportunities. At its core, CLARE leverages structured learning categories as semantic anchors to align user intent, educational taxonomies, and content retrieval. The framework comprises two key modules: (1) a BERT4Rec-based model that forecasts the next relevant learning categories from a learner’s historical interactions and (2) a category-aware prompt engineering strategy enhanced by In-Context Learning (ICL) that supplies the LLM with representative learning patterns to generate highly personalized learning recommendations. We evaluate CLARE on a real-world training dataset under both full-context and cold-start conditions. Our results demonstrate that integrating category-driven retrieval and ICL-augmented prompt design substantially improves recommendation relevance and coverage. These findings underscore the promise of combining LLM-powered RAG pipelines with domain-structured metadata to deliver adaptive, scalable, and pedagogically coherent learning guidance.