Abstract: Tracking students' historical interactions to predict future performance is a critical research focus in the field of Knowledge Tracing (KT). Recent KT approaches aim to explore students' learning patterns and evolving knowledge states to provide personalized learning guidance but often overlook the richness of exercises themselves. Additionally, with the emergence of new disciplines and interdisciplinary fields, Graph Neural Network (GNN)-based KT methods face inherent challenges, such as broadening the scope of concept associations and modeling students' learning behaviors. To address these challenges, a novel KT model, Concept association memory network by fusing graph attention knowledge tracking (GTMKT), was proposed. GTMKT models students' exercise records to trace their knowledge states and captures the global features of relevant concepts from the exercise-concept graph. Forgetting gates and higher-order information extraction were incorporated into the model to realistically simulate students' exercise processes. Through extensive experiments, compared with seven models, including Graph-based Knowledge Tracing (GKT) and Memory-Enhanced Knowledge Tracing (DGMN), on the Junyi, ASSIST09, and Static2011 datasets, the AUC and ACC performance improved by approximately 2.1% and 2.4%, respectively, demonstrating that GTMKT outperforms the baseline models on well-established knowledge structure datasets.

Key words: Knowledge Tracing&#40, KT&#41, graph neural network, self attention

摘要： 追踪学生的历史互动以预测其未来表现是知识追踪(KT)领域的关键研究重点。最近的知识追踪方法旨在研究学生的学习模式和不断变化的知识状态，以提供个性化学习指导，但忽略了习题本身的丰富性。此外，随着新专业和跨学科领域的兴起，基于图神经网络 (GNN) 的KT方法面临一些固有挑战：扩展概念之间关联的视野并建模学生的学习行为。为了解决这些挑战，提出了一种新的知识追踪模型——融合图注意力的概念关联记忆网络知识追踪 (GTMKT)。利用GTMKT，对学生的习题记录进行了建模，以追踪他们的知识状态，从习题-概念图中捕捉相关概念的全局特征。同时，将遗忘门和高阶信息提取纳入模型，以真实模拟学生的习题过程。通过广泛的实验，与基于图神经网络的结构知识追踪(GKT)和记忆增强知识追踪(DGMN)等7种模型在Junyi、ASSIST09和Static2011数据集的性能对比上，AUC和ACC性能分别提升了约2.1%和2.4%，证明GTMKT在成熟的知识结构数据集上优于对比模型。

关键词: 知识追踪, 图神经网络, 自注意力, 概念关联, 遗忘机制