基于多特征双塔结构的知识追踪模型  

Multi-feature Two-tower Structure Based Knowledge Tracing

作　　者：杨丽 

机构地区：[1]贵州民族大学数据科学与信息工程学院,贵州贵阳550025

出　　处：《工业控制计算机》2024年第8期88-90,共3页Industrial Control Computer

摘　　要：知识追踪(KT)旨在根据学习者先前的学习记录来预测他们未来的学习状态。深度知识追踪(DKT)是近年来利用深度学习技术发展起来的一种学习方法,动态跟踪学生学习情况以提供个性化的学习支持。然而,当前的研究旨在研究练习和知识点之间的关系,忽略了学生个体和知识本体的复杂互动关系。为了解决该问题,提出一种新颖的多特征双塔结构的知识跟踪模型(Multi-feature Two-tower Structure based Knowledge Tracing,MTKT),利用双塔结构中的学生塔和知识塔来相互影响学生学习的整个过程。MTKT框架应用Transformer神经网络将众多特征序列映射为多特征矩阵,同时注重不同学生个体的特征和文本信息的依赖关系的影响。在两个公开数据集上实验,结果表明,该方法比几种基线具有更好的预测性能。Deep knowledge tracing(DKT)is a learning method developed by using deep learning technology in recent years,which dynamically traces students'learning situation to provide personalized learning support.However,the current research aims to study the relationship between exercises and knowledge points,ignoring the complex interaction between individual students and knowledge ontology.In order to solve this problem,this paper proposes a novel multi-feature two-tower structure based knowledge tracing(MTKT)model,which uses the student tower and knowledge tower in the two-tower structure to influence the whole process of student learning.MTKT framework applies Transformer neural network to map many feature sequences into multi-feature matrices,while paying attention to the influence of the characteristics of different student individuals and the dependence of text information.Experiment on two public datasets results show that this method has better prediction performance than several baselines.

关 键 词：知识追踪 深度知识追踪 个性化学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]