基于深度学习的专业课程知识图谱研究

Research on Knowledge Graph for Professional Courses Based on Deep Learning

摘要: 专业课程知识体系具有强逻辑性、层级化关联与跨模块依赖特征，传统线性教材、碎片化课件难以捕捉知识点隐性语义与动态逻辑依赖；现有知识图谱(Knowledge Graph, KG)方法多聚焦通用场景，缺乏对专业课程学科逻辑约束(如先修关系不可逆)与教育语义需求(如学习路径时序性)的适配，限制其在智能教学中的应用。本文针对专业课程KG的表示学习的核心问题，从表示空间(双曲流形建模层级结构、复向量空间捕捉关系方向)、评分函数(距离基适配先修/支撑等顺序关系)、编码模型(GNN捕捉知识点局部结构、Transformer实现跨课程长距离关联)、辅助信息(融合文本/视觉/类型多模态资源)四维度，构建适配课程场景的深度学习方法体系。

Abstract: The knowledge system of professional courses is characterized by strong logicality, hierarchical relationships, and cross-module dependencies. Traditional linear textbooks and fragmented course materials struggle to capture the implicit semantics and dynamic logical dependencies among knowledge points. Existing Knowledge Graph (KG) methods mostly focus on general scenarios and lack adaptability to the disciplinary logic constraints (e.g., the irreversibility of prerequisite relationships) and educational semantic requirements (e.g., the temporal nature of learning paths) of professional courses, which limit their application in intelligent teaching. This paper addresses the core issue of representation learning for professional course KGs by proposing a deep learning framework tailored to the course context. The framework is developed from four perspectives: representation space (hyperbolic manifolds for modeling hierarchical structures, complex vector spaces for capturing relationship directions), scoring functions (distance-based adaptation for sequential relationships such as prerequisites and supports), encoding models (GNNs for capturing local structures of knowledge points, Transformers for long-range dependencies across courses), and auxiliary information (integrating multimodal resources such as text, visuals, and types).