| 摘 要: 针对现有学生知识追踪方法难以同时处理静态知识依赖与动态认知演变的问题,提出一种基于多视角图融合的学生深度知识追踪方法。首先,构建多视角图表示提取模块,利用异构图挖掘学生、项目与技能的静态语义关联,同时基于会话图捕捉学生知识状态随时间的动态演化规律。然后,设计跨视角自适应交互模块,引入交叉注意力实现学生认知状态与目标项目特性的深度对齐,并结合基于时间间隔的自适应遗忘门控,显式建模学习中的知识留存与遗忘行为。最后,在二个数据集上的实验表明本文方法在三个指标上均优于主流方法,证明本文方法的优越性。 |
| 关键词: 知识追踪 多视角图表示学习 注意力融合 自适应遗忘门控 |
|
中图分类号: TP391
文献标识码:
|
| 基金项目: 2024 年度山东省职业教育教学改革研究项目《新质生产力背景下高职信息类创新型拔尖技术人才培养模式研究与实践》,项目编号：2024096 |
|
| A Deep Knowledge Tracing Method for Students Based on Multi-view Graph Fusion |
|
Jinfeng Wang, Weirong Zhang, Xiwei Xu
|
School of Information Engineering, Weifang Vocational College
|
| Abstract: To address the limitations of existing knowledge tracing methods in simultaneously modeling long-range dependencies of static knowledge structures and the dynamic evolution of student cognitive states, this paper proposes a student deep knowledge tracing method based on multi-view graph fusion. First, a multi-view graph representation extraction module is constructed, utilizing heterogeneous graphs to mine the static semantic correlations among students, exercises, and skills, while simultaneously employing session-based graphs to precisely capture the dynamic evolution patterns of students" knowledge states over time. Subsequently, a cross-view adaptive interaction module is designed, which introduces a cross-attention mechanism to achieve deep alignment between student cognitive states and target exercise characteristics. Furthermore, an adaptive forgetting gate based on time intervals is integrated to explicitly model the mechanisms of knowledge retention and decay during the learning process. Experiments on two public datasets demonstrate that the proposed method consistently outperforms state-of-the-art baselines across three key evaluation metrics, validating the superiority of the proposed framework. |
| Keywords: Knowledge tracing multi-view graph representation learning attention fusion adaptive forgetting gate. |
