MDKT:融入多维问题难度的自适应知识追踪模型

MDKT: adaptive knowledge tracing model incorporatingmultidimensional problem difficulty

知识追踪旨在评估学习者的知识掌握状态,然而已有研究表明,问题难度与知识掌握状态密切相关。忽略问题难度的知识追踪模型难以有效评估学习者的实际状态。为了解决上述问题,提出了融入多维问题难度的自适应知识追踪模型(multi-dimensional knowledge tracing, MDKT)。该模型采用BERT与CNN对题目文本进行语义难度提取,并结合问题难度、概念难度和认知难度,形成多维问题难度表征;通过构建自适应学习模块,个性化地捕捉学习者与增强练习难度之间的交互;在预测学习者未来表现过程中,引入Transformer的多头注意力机制,以关注不同部分预测状态的重要程度。在实验阶段,与七个知识追踪模型在两个真实数据集的性能对比实验中,AUC、ACC性能分别提升了3.99%~12.06%和3.63%~11.15%,实验结果表明,所提模型在性能方面更加出色。在应用方面,将该模型和知识点网络图相结合,能准确挖掘出学习者的薄弱知识点,证明了所提模型在实际教学中的可行性。

Knowledge tracing aims to assess learners’mastery of knowledge,but studies have shown that question difficulty is closely related to mastery status.Models that overlook question difficulty struggle to effectively evaluate learners’actual status.To resolve this issue,this paper developed an MDKT model,incorporating multi-dimensional difficulty.This model employed BERT and CNN to extract semantic difficulty from question texts and integrates question difficulty,conceptual difficulty,and cognitive difficulty to create a multi-dimensional difficulty representation.It constructed an adaptive learning module to capture the interaction between learners and increased exercise difficulty personally.In predicting learners’future perfor-mance,the model used the Transformer’s multi-head attention mechanism to focus on the importance of different prediction states.Experimentally,on two real datasets,the MDKT model improved performance by 3.99%~12.06%in AUC and 3.63%~11.15%in ACC,outperforming seven other knowledge tracing models.The results demonstrate the superior performance of the model.Furthermore,integrating this model with a knowledge point network graph accurately identifies lear-ners’weak knowledge points,and it confirms the model’s feasibility in actual teaching.