Scholarlink

<p>Knowledge tracing (KT) models students’ knowledge states to predict future performance based on historical interactions. Due to data privacy concerns and budget constraints, the availability of high-quality student data differs across domains and it is essential to effectively utilize KT data from multiple domains. In this work, we propose a novel prompt-enhanced paradigm, i.e., <i>promptKT</i>, to utilize student data from multiple domains to improve KT performance simultaneously. Specifically, a unified Transformer based backbone model is first pre-trained using data from all the KT domains to capture the commonality across domains. Then, we design a novel soft domain prompt module to capture the distinctions among various domains and users. Our promptKT is evaluated on six public real-world educational datasets. The results demonstrate that our approach outperforms the majority of existing KT models in terms of AUC and accuracy. Furthermore, empirical analysis shows the decent transferability and adaptation of promptKT across multiple KT domains. To encourage reproducible research, we make our data and code publicly available at <a href="https://github.com/pykt-team/pykt-toolkit">https://github.com/pykt-team/pykt-toolkit</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A prompt-driven framework for multi-domain knowledge tracing

  • 期刊论文

  • 发表日期：

    2025年2月17日

查看全文
  • Zitao Liu,
  • Shuyan Huang,
  • Teng Guo,
  • Mingliang Hou,
  • Qianru Liang

摘要

Knowledge tracing (KT) models students’ knowledge states to predict future performance based on historical interactions. Due to data privacy concerns and budget constraints, the availability of high-quality student data differs across domains and it is essential to effectively utilize KT data from multiple domains. In this work, we propose a novel prompt-enhanced paradigm, i.e., promptKT, to utilize student data from multiple domains to improve KT performance simultaneously. Specifically, a unified Transformer based backbone model is first pre-trained using data from all the KT domains to capture the commonality across domains. Then, we design a novel soft domain prompt module to capture the distinctions among various domains and users. Our promptKT is evaluated on six public real-world educational datasets. The results demonstrate that our approach outperforms the majority of existing KT models in terms of AUC and accuracy. Furthermore, empirical analysis shows the decent transferability and adaptation of promptKT across multiple KT domains. To encourage reproducible research, we make our data and code publicly available at https://github.com/pykt-team/pykt-toolkit.