多关系和时间增强的知识追踪模型  

Multiple relations and time enhanced knowledge tracing model

作　　者：张维[1] 罗佩华 龚中伟 李志新 宋玲玲[1] Zhang Wei;Luo Peihua;Gong Zhongwei;Li Zhixin;Song Lingling(Faculty of Artificial Intelligence Education,Central China Normal University,Wuhan 430079,China)

机构地区：[1]华中师范大学人工智能教育学部,武汉430079

出　　处：《计算机应用研究》2025年第3期728-734,共7页Application Research of Computers

基　　金：国家自然科学基金面上项目(62377024)。

摘　　要：现有知识追踪方法未能深入探索知识点间多种关系并同时考虑知识相互作用和时间对知识状态的影响。鉴于此,从知识间多种关系和学习遗忘规律两方面改进知识追踪模型,提出多关系和时间增强的知识追踪模型(MRTKT)。首先,根据认知同化理论丰富了知识间关系,使用统计学方法构建了包含上位学习、下位学习以及并列组合学习三种关系的知识结构图;其次,对知识间相互作用进行建模,根据上述三种关系聚集节点特征,使得模型可以更好地模拟知识间的影响传播行为;然后,构建融入三种时间信息的GRU门更新学生知识状态,以模拟学习和遗忘对知识状态的影响,使得各节点特征包含知识间相互作用信息和时间信息,为预测学习者答题表现提供更全面丰富的信息。在多个公开数据集上进行实验,结果表明MRTKT比现有模型具有更优越的性能以及更好的可解释性。Existing knowledge tracing methods fail to effectively explore and utilize the multiple relations between concepts and simultaneously consider the effects of interactions between concepts as well as time on the knowledge state.This paper improved the knowledge tracing model in terms of multiple relations between concepts and learning-forgetting patterns,and proposed a multiple relations and time enhanced knowledge tracing model(MRTKT).Firstly,it enriched the relations between concepts according to assimilation theory,and constructed a knowledge structure containing three relationships of superordinate learning,subordinate learning,and combinatorial learning by using a statistical methodology.Secondly,it modeled the inte-raction between concepts enables the aggregation of node features based on the above three relationships.This enabled the model to better simulate influence propagation among concepts.Then,it updated knowledge states using a gate mechanism incorporating three temporal factors in order to simulate the learning-forgetting effect.This ensures that each node feature contains both interactions between concepts and time information,providing more comprehensive and rich information for predicting learners’responses.It conducted experiments on three real-world datasets,and the results show that MRTKT has superior performance and better interpretability than existing models.

关 键 词：知识追踪 门控图神经网络 GRU 知识间多种关系 学习遗忘 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]