知识图谱协同大模型的个性化学习路径推荐框架构建与教学实践探索
Construction and Teaching Practice Exploration of a Personalized Learning Path Recommendation Framework Synergizing Knowledge Graphs and Large Language Models
王育恒 1郭一珺1
作者信息
- 1. 北京邮电大学信息与通信工程学院,北京 100876;先进信息网络北京市重点实验室（北京邮电大学），北京 100876
- 折叠
摘要
在通信原理等复杂工程学科教学中,知识点的高抽象性与强逻辑关联常导致学生认知负荷过高,而传统“一刀切”模式难以满足差异化需求。现有适应性学习技术面临严重的数据依赖与“冷启动”困境,生成式人工智能虽交互灵活,却存在“知识幻觉”与缺乏教学法约束的风险。为此,本文构建了一种知识图谱协同大语言模型的路径推荐框架,该研究以领域知识图谱为“教学认知导航图”,利用其结构化属性对大模型推理施加“刚性”约束,通过评估三元组与前驱关系约束,解决了大模型生成内容的随机性问题。在此基础上,设计了涵盖“动态学情评估—知识薄弱点诊断—学习者画像构建—个性化路径规划”的自动化闭环算法,并开发KG-Tutor系统应用于《通信原理》教学。实证表明,该系统有效解决了无历史数据下的落地难题；相较于对照组,实验组学生的知识掌握度在7周后提升了\qty{24.5}{\percent},且生成的路径在逻辑性与准确性上表现优异。本研究为构建新一代可靠、可控且无需预热的学习路径推荐系统提供了实践蓝图,对深化教育人工智能应用具有重要参考价值。
Abstract
In the teaching of complex engineering disciplines like \textit{Communication Principles}, the high abstraction and strong logical correlation of knowledge points often lead to excessive cognitive load, while traditional ``one-size-fits-all'' modes fail to meet differentiated needs. Existing adaptive learning technologies face severe data dependencies and ``cold start'' dilemmas, while generative AI, though flexible in interaction, carries risks of ``knowledge hallucinations'' and a lack of pedagogical constraints. To address these issues, this paper constructs a path recommendation framework synergizing Knowledge Graphs (KG) with Large Language Models (LLM). Using the domain KG as a ``pedagogical cognitive map,'' the framework imposes pedagogical constraints on LLM reasoning via structured attributes, specifically addressing the randomness of LLM generation through evaluation triples and precursor relationship constraints. Furthermore, an automated closed-loop algorithm covering ``dynamic learning assessment, knowledge weakness diagnosis, learner profiling, and personalized path planning'' was designed, and the KG-Tutor system was developed and applied to the teaching of \textit{Communication Principles}. Empirical results show that the system effectively solves deployment challenges in the absence of historical data; compared to the control group, the experimental group achieved a \qty{24.5}{\percent} improvement in knowledge mastery over the 7-week period, with the generated paths demonstrating superior logic and accuracy. This study provides a practical blueprint for constructing a new generation of reliable, controllable, and cold-start resilient learning path recommendation systems, offering significant reference value for deepening the application of Educational Artificial Intelligence (EAI).关键词
计算机应用/学习路径推荐系统/教育人工智能/适应性学习/个性化学习路径/知识图谱Key words
Computer Applications/Learning Path Recommendation System/Educational Artificial Intelligence (EAI)/Adaptive Learning/Personalized Learning Path/Knowledge Graph引用本文复制引用
王育恒,郭一珺.知识图谱协同大模型的个性化学习路径推荐框架构建与教学实践探索[EB/OL].(2026-01-21)[2026-01-22].http://www.paper.edu.cn/releasepaper/content/202601-33.学科分类
教育/计算技术、计算机技术
评论