Abstract: Existing knowledge tracing (KT) models often fail to effectively leverage learning behavior information and tend to overlook the varying contributions of different learning behaviors to students' performance. To address these limitations, this paper proposes a Learning behavior boosted knowledge tracing (LBBKT). The model employs a Gated Residual Network (GRN) to encode student learning behavior features into four contextual vectors, which are then integrated into the model. This approach fully utilizes learning behavior data, such as response speed, number of attempts, and hints, to better model the learning process. Additionally, the model applies a variable selection network to selectively weight student learning behavior features, while the GRN further suppresses irrelevant features, thereby amplifying the impact of relevant features on students' performance. Experimental results on multiple public datasets demonstrate that the LBBKT model significantly outperforms existing KT models in terms of prediction accuracy.

Key words: knowledge tracing, learning behavior, variable selection network, selective weighting, gated residual network

摘要： 现有的知识追踪(KT)模型无法有效利用学习行为信息，且忽略了不同学习行为对答题表现的贡献不同。为此，本文提出了一种学习行为增强的知识追踪模型(LBBKT)。该模型采用门控残差网络(GRN)将学生学习行为特征编码成四种上下文向量融入模型中，充分利用了学习行为信息(答题速度，尝试次数，提示)，更好地建模学生学习过程。此外，模型利用变量选择网络对学生的学习行为特征进行选择性加权，并利用GRN进一步抑制不必要的特征，增强相关特征对学生答题表现的影响。最后，在多个公开数据集上的实验结果表明，学习行为增强的知识追踪模型在预测准确性上显著优于现有模型。

关键词: 知识追踪, 学习行为, 变量选择网络, 选择性加权, 门控残差网络