Zhou Xin, Zhang Zhuoxu, Xie Xike, Zhang Jiawei
State University of New York at Binghamton, Binghamton, New York, USA.
The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China.
Sci Rep. 2025 Jul 1;15(1):21395. doi: 10.1038/s41598-025-07422-7.
The emergence of online education, e.g., intelligent tutoring system (ITS), complements or partially replaces conventional offline education, especially during the COVID-19 pandemic. Knowledge tracing (KT) plays a pivotal role in the intelligent tutoring system in capturing the knowledge states of students. By analyzing a series of students' interaction records of questions and answers in ITS, KT is able to provide personalized feedbacks to students. Recent advances in deep learning techniques, such as deep knowledge tracing, apply recurrent neural networks over students' interaction records for knowledge state modeling and achieve great improvement in the prediction of performance on future tasks and assessment questions. However, in practice, KT is often in lack of sufficient student interaction records to accurately model and predict students' knowledge states, the so-called data sparsity issue. Meanwhile, the data sparsity issue is generally overlooked in the existing knowledge tracing systems. In this paper, we propose a quality-aware deep learning framework for knowledge tracing, based on the sparse attention techniques and generative decoding. Extensive experiments are conducted over a series of real datasets showing that our proposal accurately captures students' knowledge states.
在线教育的出现，例如智能辅导系统（ITS），补充或部分取代了传统的线下教育，尤其是在新冠疫情期间。知识追踪（KT）在智能辅导系统中对于捕捉学生的知识状态起着关键作用。通过分析智能辅导系统中一系列学生的问答互动记录，知识追踪能够为学生提供个性化反馈。深度学习技术的最新进展，如深度知识追踪，将循环神经网络应用于学生的互动记录以进行知识状态建模，并在预测未来任务和评估问题的表现方面取得了巨大进步。然而，在实践中，知识追踪往往缺乏足够的学生互动记录来准确建模和预测学生的知识状态，即所谓的数据稀疏问题。同时，现有知识追踪系统普遍忽视了数据稀疏问题。在本文中，我们基于稀疏注意力技术和生成式解码提出了一种用于知识追踪的质量感知深度学习框架。在一系列真实数据集上进行了广泛实验，结果表明我们的方案能够准确捕捉学生的知识状态。
