融合学习过程和学习行为的深度知识追踪模型  

Deep Knowledge Tracing Model Integrating Learning Process and Behavior

作　　者：贾瑞 董永权[1,2,3] 陈成 刘源 JIA Rui;DONG Yongquan;CHEN Cheng;LIU Yuan(School of Computer Science and Technology,Jiangsu Normal University,Xuzhou 221116,China;Jiangsu Education Informatization Engineering Technology Research Center,Xuzhou 221116,China;Xuzhou Cloud Computing Engineering Technology Research Center,Xuzhou 221116,China)

机构地区：[1]江苏师范大学计算机科学与技术学院,江苏徐州221116 [2]江苏省教育信息化工程技术研究中心,江苏徐州221116 [3]徐州云计算工程技术研究中心,江苏徐州221116

出　　处：《哈尔滨理工大学学报》2024年第5期18-28,共11页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金(61872168);江苏省教育科学“十四五”规划项目(D/2021/01/112);江苏师范大学研究生科研与实践创新项目(2022XKT1549)。

摘　　要：知识追踪的目的是持续评估学生的知识状态,并预测学生未来的学习表现。学生的学习过程本质上是学生、知识点和习题之间的相互作用,学生通过学习行为影响学习过程。学习行为包括获取知识行为和遗忘知识行为。为了准确建模知识追踪中的学习过程和学习行为,提出一种融合学习过程和学习行为的深度知识追踪模型。该模型结合项目反应理论和LSTM建模知识点与习题之间的相互作用,利用单调注意力机制拟合学生的学习行为,并且定义两个解码器用以捕获学生与知识点、习题之间的相互作用,从而融合学习过程和学习行为。在真实数据集ASSISTment2009和ASSISTment2017上的实验结果表明,该模型的表现优于已有的知识追踪模型,相比于次优模型,模型的预测准确率在两个数据集上均有1%的提高。The purpose of knowledge tracking is to continuously assess the state of students′knowledge and to predict their future learning performance.The student learning process is essentially an interaction between students,knowledge points,and exercises,and students influence the learning process through their learning behaviors.Learning behaviors include knowledge acquisition behaviors and knowledge forgetting behaviors.In order to accurately model the learning process and learning behavior in knowledge tracking,a deep knowledge tracking model that integrates the learning process and learning behavior is proposed.The model combines item response theory and LSTM to model the interactions between knowledge points and exercises,uses a monotonic attention mechanism to fit students′learning behaviors,and defines two decoders to capture the interactions between students and knowledge points and exercises,thus fusing learning processes and learning behaviors.Experimental results on the real datasets ASSISTment2009 and ASSISTment2017 show that the model outperforms existing knowledge tracking models,and the prediction accuracy of the model is improved by 1%on both datasets compared to the suboptimal model.

关 键 词：知识追踪 项目反应理论 学习行为 深度学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]