融入动态技能学习时间分类的知识追踪模型  

Knowledge Tracing Model Incorporating Dynamic Skill Learning Time Classification

作　　者：宋刚 许晓东[1] SONG Gang;XU Xiaodong(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212000)

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212000

出　　处：《计算机与数字工程》2022年第4期849-854,共6页Computer & Digital Engineering

摘　　要：为解决当前DKT模型忽略学生实际答题时间带来的个体化差异问题,通过融入学生每个技能前后两次的答题时间,对学生技能学习的时间进行动态分组。同时,引入与原始DKT模型的损失函数相对应的重构正则化项,寻找学生答题时间个体化差异带来影响的最优解。在ASSISTments和Cognitive Tutor数据集上的对比实验证实AUC最高提升了10%的收益。In order to solve the problem of individual differences caused by the current deep knowledge tracing(DKT)model ignoring students’actual answering time,the students’skills learning time is dynamically grouped by integrating the students’answering time before and after each skill.At the same time,the reconstruction regularization term corresponding to the loss function of the original DKT model is introduced to find the optimal solution that is affected by the individual differences in student answer time.Comparative experiments on the ASSISTments and Cognitive Tutor datasets confirm that the AUC improves the revenue by up to 10%.

关 键 词：深度学习 个性化学习 知识追踪 数据挖掘 

分 类 号：G434[文化科学—教育学]