DKTwMF:一种融合多特征的知识追踪模型  

DKTwMF:A Deep Knowledge Tracing Model with Multiple Features

作　　者：王斌[1] 盛宇轩 冀星昀 WANG Bin;SHENG Yu-xuan;JI Xing-yun(School of Computer Science and Engineering,Central South University,Changsha 410083,China)

机构地区：[1]中南大学计算机学院,湖南长沙410083

出　　处：《计算机技术与发展》2021年第7期35-41,共7页Computer Technology and Development

基　　金：国家科技重大专项项目(2017ZX06002005)。

摘　　要：知识追踪的目的是通过分析学习活动来量化学生掌握知识的能力,进而为每个学生提供更具有针对性的训练。伴随时间的推移,系统中记录的学习数据的数据量不断增加。如何充分利用这些数据,为学生提供个性化教育是目前教育数据挖掘领域的一个重要研究方向。现有的知识追踪模型大多只考虑学生练习相关的知识点和作答结果,并未充分利用数据集中其他数据。为解决上述问题,提出了一种融合多特征的知识追踪模型(deep knowledge tracing with multiple features, DKTwMF)。该模型首先利用邻域互信息和随机森林对特征进行选择,然后将多特征数据进行编码,最后使用深度学习对学生知识掌握状态进行建模。该模型可以自动提取出数据集中的重要特征,减少训练参数,保证模型能够收敛到全局最优,从而更准确地对学生技能状态进行评估。The purpose of knowledge tracking is to quantify the students’ability of grasping knowledge by analyzing learning activities,and then provide more targeted training for each student.With the passage of time,the learning data recorded in this system is increasing.How to make full use of these data to provide students with personalized education is an important research direction in the field of educational data mining.Most of the existing models only consider knowledge points and answers related to student practice and do not make full use of other data in the data set.In order to solve the above problems,we present a multi-feature knowledge tracking model,DKTwMF.Firstly,the model uses the neighborhood mutual information and random forest to select the features,and then encodes the multi feature data.Finally,deep learning is used to model student knowledge.This model can automatically extract the important features of the data set,reduce the number of training parameters,and ensure the model can converge to the global optimal,so as to evaluate the students’skill state more accurately.

关 键 词：数据挖掘 知识追踪 教育系统 特征融合 深度学习 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]