可信的端到端深度学生知识画像建模方法  

Trustworthy End-to-End Deep Student Knowledge Portrait Modelling Method

作　　者：王士进 吴金泽 张浩天 沙晶 黄振亚 刘淇[3] Wang Shijin;Wu Jinze;Zhang Haotian;Sha Jing;Huang Zhenya;Liu Qi(State Key Laboratory of Cognitive Intelligence,iFLYTEK Co.,Ltd,Hefei 230088;iFLYTEK Co.,Ltd,Hefei 230088;Anhui Province Key Laboratory of Big Data Analysis and Application,University of Science and Technology of China.Hefei 230027)

机构地区：[1]认知智能全国重点实验室(科大讯飞股份有限公司),合肥230088 [2]科大讯飞股份有限公司,合肥230088 [3]大数据分析与应用安徽省重点实验室(中国科学技术大学),合肥230027

出　　处：《计算机研究与发展》2023年第8期1822-1833,共12页Journal of Computer Research and Development

基　　金：国家重点研发计划(2021YFF0901003)。

摘　　要：学生知识画像是对学生在不同知识概念掌握程度的全面精准的表示.通常,智能教育系统中使用知识追踪方法,基于显式的学生交互数据,对学生在某些知识概念的隐式掌握程度进行建模.然而知识追踪方法的预测结果与学生知识画像存在着时序、预测粒度不一致的情况,导致其产生的学生知识画像不可信.对此,首先基于端到端的学生知识掌握度预测目标定义并形式化学生知识画像预测任务,然后提出了一种深度知识画像(deep knowledge portrait, DKP)模型.该方法首先在知识粒度上学习交互表征,引入了知识难度、知识概念等特征在知识粒度上区分交互;然后,采用双向长短时记忆网络基于学生历史交互序列,建模学生知识状态变化.最后针对待预测知识概念,使用了多头注意力池化层强化历史序列中的相关交互以进行该概念下的学生掌握度预测.在3个真实的数据集上的实验结果表明,所提出的方法更适合学生知识画像预测任务从而获得更可信的学生知识画像,并在各项性能上超过了现有的方法.Student knowledge portrait means a comprehensive and accurate representation of student’s mastery for knowledge concepts.Generally,knowledge tracing methods are used in intelligent education to summarize and predict potential students’mastery of knowledge concepts based on students’learning data.However,the prediction of the knowledge tracing method is inconsistent with the student knowledge portrait,leading to not credible portrait.In this paper,we first define the student knowledge portrait task based on the end-to-end object of students’knowledge mastery,and then propose a novel deep knowledge portrait(DKP)model.Specifically,we first represent the learning interaction with difficulty and knowledge concepts to distinguish learning interactions on the knowledge level.Besides,we adopt the bidirectional long and short time memory network(Bi-LSTM)to trace the student’s knowledge states based on the learning record.Finally,we predicted the student knowledge portrait using a multi-head attention pooling layer to focus on the historical knowledge states related to the knowledge concepts to predict the mastery.Experimental results on three real datasets show that the proposed method is more suitable for student knowledge portrait task so as to obtain a more trustworthy student knowledge portrait,and overperforms the baselines on metrics.

关 键 词：学生知识画像 知识追踪 智能教育 数据挖掘 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]