Scholarlink

<p>In zero-shot scenarios where learners lack historical interaction data, existing personalized curriculum recommendation methods often struggle to generate accurate recommendations and offer limited interpretability. In this work, we propose Learning Like a Human, a novel framework for zero-shot personalized course recommendation, which combines cognitive diagnosis modeling with LLM agents to simulate human-like learning processes. The proposed framework consists of three key components: (1) a cognitive diagnosis-based learner profile module that dynamically represents a learner’s knowledge state; (2) a personalized learning based course path recommendation module, which leverages user features, cognitive embeddings, and action patterns, to generate adaptive course sequences; and (3) an LLM-powered action simulation agent, which models cognitive responses and actions for cold-start users, enabling the system to incremental learn from synthetic yet realistic trajectories. Extensive experiments on multiple real-world educational datasets show that our approach significantly outperforms state-of-the-art baselines in both recommendation precision and average learner performance. The recall rate and precision improvement are as high as 4.9% and 3.4%, respectively. Moreover, unlike existing static interpretable systems that rely heavily on prior knowledge, the proposed method provides instructional explanations based on real-time cognitive diagnosis, with each recommendation step accompanied by explicit cognitive and behavioral rationale. The code will be released at GitHub.</p>

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

Learning Like a Human: Zero-shot Personalized Course Recommendation System via Cognitive Diagnosis and LLM Agents

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月7日

查看全文
  • Wen Fu

摘要

In zero-shot scenarios where learners lack historical interaction data, existing personalized curriculum recommendation methods often struggle to generate accurate recommendations and offer limited interpretability. In this work, we propose Learning Like a Human, a novel framework for zero-shot personalized course recommendation, which combines cognitive diagnosis modeling with LLM agents to simulate human-like learning processes. The proposed framework consists of three key components: (1) a cognitive diagnosis-based learner profile module that dynamically represents a learner’s knowledge state; (2) a personalized learning based course path recommendation module, which leverages user features, cognitive embeddings, and action patterns, to generate adaptive course sequences; and (3) an LLM-powered action simulation agent, which models cognitive responses and actions for cold-start users, enabling the system to incremental learn from synthetic yet realistic trajectories. Extensive experiments on multiple real-world educational datasets show that our approach significantly outperforms state-of-the-art baselines in both recommendation precision and average learner performance. The recall rate and precision improvement are as high as 4.9% and 3.4%, respectively. Moreover, unlike existing static interpretable systems that rely heavily on prior knowledge, the proposed method provides instructional explanations based on real-time cognitive diagnosis, with each recommendation step accompanied by explicit cognitive and behavioral rationale. The code will be released at GitHub.