New Knowledge Tracking Model Outperforms Baselines in Predicting Student Performance

A new knowledge tracking model designated as PKS-CLB achieves superior predictive performance over state-of-the-art baselines when evaluated on a real-world educational dataset, demonstrating its effectiveness in advancing personalized learning support. Developed to align potential knowledge states with learning gain consistency, the framework predicts learner performance on upcoming test questions within intelligent tutoring systems.

Knowledge tracking serves as a foundational technology for intelligent tutoring systems by modeling how learners’ mastery of specific knowledge points evolves over time. Existing computational approaches face notable limitations, particularly an insufficient exploration of latent knowledge states, inconsistent handling of learning gains, and the neglect of correlations between different knowledge points.

To overcome these limitations, the PKS-CLB model incorporates specialized mechanisms to capture and infer learner progression. The architecture employs a multi-head attention mechanism structured to capture improvements in knowledge states following both correct and incorrect learner responses, while simultaneously utilizing a knowledge graph to infer learners’ latent knowledge states.

Following this dual processing, a bidirectional interaction module produces a unified final representation of the learner’s knowledge state. This synthesized representation is then applied directly to predict performance on the next test question, integrating potential states and learning benefit consistency into a coherent tracking framework.

Subject of Research: Knowledge tracking in intelligent tutoring systems

Article Title: A knowledge tracking model based on the potential knowledge states and consistency of learning benefits

Article References: Luo, X., Wang, R., & Meng, Y. (2026). A knowledge tracking model based on the potential knowledge states and consistency of learning benefits. Knowledge and Information Systems, 68(1), Article 315. https://doi.org/10.1007/s10115-026-02940-w

Image Credits: AI Generated

DOI: 10.1007/s10115-026-02940-w

Keywords: knowledge tracking, intelligent tutoring systems, PKS-CLB, potential knowledge states, learning gain consistency, multi-head attention, knowledge graphs, bidirectional interaction module, predictive performance, personalized learning support, latent knowledge states, inter-knowledge-point correlations, test performance prediction, educational datasets, learning gains, student modeling