What question did this study set out to answer?

This research aims to optimize personalized learning paths in digital English environments by integrating knowledge tracing with reinforcement learning.

May 29, 2026Open Access

Personalised learning path optimisation in digital English learning environments via multi-factor knowledge tracing and reinforcement learning

Key Points

This research aims to optimize personalized learning paths in digital English environments by integrating knowledge tracing with reinforcement learning.
Proposed a multi-factor fusion knowledge tracing model that analyzes behavior sequences and item attributes.
Conducted experiments using ASSISTments 2017 and EdNet-KT4 datasets to validate model performance.
Utilized conservative Q-learning for enhanced path optimization and compared it with rule-based and random strategies.
MFFKT achieved AUC scores of 0.834 and 0.812, outperforming baseline models.
The method showed superior performance in cumulative reward and efficiency compared to traditional strategies.

Abstract

Digital English learning environments generate massive interaction data, offering potential for adaptive learning path optimisation.However, many existing approaches treat learning state estimation and learning path recommendation as separate tasks, restricting long-term personalised learning support.This study proposes a framework that integrates a multi-factor fusion knowledge tracing (MFFKT) model with reinforcement learning.It jointly analyses behaviour sequences, item attributes, knowledge structures and temporal features to dynamically capture learners' knowledge states, which serve as environment states for long-term reward-driven path optimisation.Experiments on ASSISTments 2017 and EdNet-KT4 show that MFFKT achieves AUC scores of 0.834 and 0.812, surpassing baseline models.Ablation studies validate the efficacy of multi-dimensional feature fusion.When combined with conservative Q-learning, the methods outperform greedy, rulebased, and random strategies in cumulative reward, completion rate, and efficiency.Overall, the proposed framework enables coordinated modelling of learning states and learning path decisions, providing an effective technical approach for adaptive and personalised English learning within digital learning environments.

Mark Helpful

Bookmark

Relay

View Full Paper