面向学习轨迹的知识追踪预测模型

A knowledge tracing prediction model for learning trajectories

下载PDF

导出

摘要基于transformer架构,提出一种面向学习轨迹的知识追踪预测模型(knowledge tracing prediction model for learning trajectories,LTKT),解决知识追踪领域使用transformer架构所存在的问题:网络中缺乏知识点信息、注意力被分散到众多关联较小的试题及忽略了学习能力在答题决策中的影响。LTKT在数据预处理阶段,采用教育领域的知识融通机制整合题目涉及的多个知识点,作为模型学习的一个信息维度。在编码器与解码器结构中,根据注意力呈现长尾分布的特点引入稀疏自注意力机制,并在其中嵌入包含绝对距离和相对距离的位置编码,使注意力集中在少数高度相似的试题上,同时加强模型对位置信息的感知。在预测策略上,使用双线性层融合学习能力特征与学生的知识状态,综合预测学生下一时刻的作答表现。在两个真实的大型公开数据集上进行实验,与其他优秀模型进行对比,结果显示LTKT的AUC有了明显提升。 Based on the transformer architecture,a knowledge tracing prediction model for learning tra‐jectory was proposed,which solved the following problems in the field of knowledge tracing using the transformer architecture:the model lacked the learning of knowledge point information;the attention scores in the self-attention mechanism showed a long-tail distribution and required square computatio-nal overhead;the prediction strategy of the model lacked consideration of learners’ability.In the data preprocessing stage,LTKT used the knowledge integration mechanism in the field of education to inte‐grate multiple knowledge points involved in the subject,and the integrated knowledge formed was used as input to the model along with other learning trajectory information;LTKT introduced a sparse self-attention mechanism according to the characteristics of the long-tail distribution of attention scores into the encoder and decoder structure,and embedded a position encoding containing absolute distance and relative distance in it,so that the deep attention mechanism could also learn the position relation‐ship between topics.In the prediction strategy,LTKT used the bilinear layer to fuse the learning ability features extracted by the learning ability extraction module and the output of the decoder to comprehen‐sively predict the student's answer performance at the next moment.Experiments were carried out on two real large public datasets,and compared with other excellent models.The results show that LTKT has significantly improved the AUC.

作者张翼飞张加金关凯俊张玉雪 ZHANG Yifei;ZHANG Jiajin;GUAN Kaijun;ZHANG Yuxue(College of Computer Science,Shenyang Aerospace University,Shenyang 110136,China)

机构地区沈阳航空航天大学计算机学院

出处《沈阳航空航天大学学报》 2024年第3期61-70,共10页 Journal of Shenyang Aerospace University

基金国家自然科学基金(项目编号:62102271)。

关键词知识追踪深度学习稀疏自注意力机制预测策略融通知识 knowledge tracing deep learning sparse self-attention mechanism forecasting strategy integration knowledge

分类号 TP391 [自动化与计算机技术—计算机应用技术]