Scholarlink

<p>Knowledge tracing (KT) aims to dynamically model learners’ evolving knowledge states based on their historical learning records, playing a vital role in online education systems. Most existing KT methods learn the knowledge states as a transition pattern from the previous exercise to the next one, treating learners’ entire learning records as continuous and uniformly distributed. However, we argue that actual learning records can be divided into distinct shorter sessions. To this end, we propose a novel KT model called Fine-grained Session Modeling for Knowledge Tracing (FSM4KT), which is designed to capture learners’ knowledge state changes with finer granularity. In particular, we first divide learners’ extensive historical learning records into shorter sessions from either temporal or knowledge concept-related perspective. Subsequently, a dedicated designed session-based knowledge proficiency modeling component is presented, which figures out intra-session and inter-session fine-grained interaction dependencies and knowledge state changes. Moreover, a global knowledge proficiency modeling component is introduced to holistically model learners’ knowledge states. Extensive experimental results on three real-world datasets demonstrate that FSM4KT outperforms most of the current baseline methods, thus proving the effectiveness of FSM4KT.</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

Enhancing knowledge tracing with fine-grained session modeling

  • 期刊论文

  • 发表日期：

    2025年1月10日

查看全文
  • Jing Wang,
  • Huifang Ma,
  • Mengyuan Zhang,
  • Zhixin Li,
  • Liang Chang

摘要

Knowledge tracing (KT) aims to dynamically model learners’ evolving knowledge states based on their historical learning records, playing a vital role in online education systems. Most existing KT methods learn the knowledge states as a transition pattern from the previous exercise to the next one, treating learners’ entire learning records as continuous and uniformly distributed. However, we argue that actual learning records can be divided into distinct shorter sessions. To this end, we propose a novel KT model called Fine-grained Session Modeling for Knowledge Tracing (FSM4KT), which is designed to capture learners’ knowledge state changes with finer granularity. In particular, we first divide learners’ extensive historical learning records into shorter sessions from either temporal or knowledge concept-related perspective. Subsequently, a dedicated designed session-based knowledge proficiency modeling component is presented, which figures out intra-session and inter-session fine-grained interaction dependencies and knowledge state changes. Moreover, a global knowledge proficiency modeling component is introduced to holistically model learners’ knowledge states. Extensive experimental results on three real-world datasets demonstrate that FSM4KT outperforms most of the current baseline methods, thus proving the effectiveness of FSM4KT.