首页|迁移表征的知识追踪模型

迁移表征的知识追踪模型

扫码查看
原文链接
  • 万方数据
  • 维普
针对多数知识追踪研究在表征题目时仅利用了题目包含的概念等显性特征,未能考虑到题目中概念的考察侧重程度这一隐性特征,也未表征迁移过程中题目的迁移程度的问题,本文提出题目迁移表征的知识追踪模型.在题目侧重表征方面,采用加性注意力机制提取题目中各个概念的考察侧重程度;在题目迁移方面,利用相似性和通道注意力机制融合建模历史题目多角度的迁移程度;在迁移遗忘方面,使用门限机制建模学习迁移的遗忘过程.最终得到题目迁移表征,以此来预测学习者未来的答题表现.在实验阶段,与 6 种相关模型在3 个真实数据集上进行对比实验,结果表明提出模型的曲线下面积(area under the curve,AUC)和准确率(accur-acy,ACC)均有更好表现,尤其在ASSISTments2012 数据集上表现最佳,相较于其他对比模型分别提升了3.5%～20.1%和 2.3%～18.5%;在可解释性方面,使用图表可视化描述了题目迁移表征生成路径.本研究建模的学习迁移内在机制可为知识追踪模型的设计提供参考.
Knowledge tracing model via exercise transfer representation
In most knowledge tracing research,the representation of questions typically relies solely on explicit features such as the concepts contained within the questions,neglecting the implicit feature of the emphasis level on concepts and failing to characterize the degree of question transfer during the transfer process.This paper presents a knowledge tracing model via exercise transfer representation.In terms of the representation of the focus of the exercise,the addit-ive attention mechanism is used to extract the focus of each concept in the exercise.In terms of exercise transfer,the multi-angle transfer degree of historical exercises is modeled by fusion of similarity and channel attention mechanism.In terms of transfer forgetting,a threshold mechanism is used to model the forgetting process of learning transfer.Fi-nally,the exercise transfer representation is obtained to predict the learner's future answering performance.Experiment-al results show that the proposed model outperforms six benchmark models across three real datasets,particularly excel-ling on the ASSISTments 2012 dataset with improvements ranging from 3.5%to 20.1%for area under the curve(AUC)and 2.3%to 18.5%for accuracy(ACC).The interpretability aspect is enhanced through graphical visualization of the question transfer representation generation pathway.The internal mechanisms of learning transfer modeled in this paper provide valuable insights for the design of knowledge tracing models.

knowledge tracinglearning transfer mechanismexercise representationexercise transfersequence modelanswer predictionattention mechanismthreshold mechanism

张凯、刘月、覃正楚、秦心怡

展开 >

长江大学 计算机科学学院,湖北 荆州 434023

知识追踪 学习迁移机制 题目表征 题目迁移 序列模型 答题预测 注意力机制 门限机制

国家自然科学基金项目国家科技部高端外国人才引进计划项目湖北省自然科学基金项目湖北省教育厅科学研究计划项目2023年湖北本科高校省级教学改革研究项目

62077018G2022027006L2022CFB132B20220382023273

2024

10.11992/tis.202302002

智能系统学报
中国人工智能学会　哈尔滨工程大学

智能系统学报

CSTPCD北大核心
影响因子：0.672
ISSN：1673-4785
年,卷(期)：2024.19(4)
  • 10