Interpretable knowledge tracing via explicit–implicit alignment

Interpretability in Knowledge Tracing (KT) is crucial for personalized learning. Current mainstream Deep Learning-based KT (DLKT) achieves strong predictive performance with ID-based representations, but students’ latent knowledge states remain opaque to teachers and learners. Large Language Models (LLMs) can generate human-readable text, but explanations from existing LLM-augmented KT approaches are either detached from the predictive state or insensitive to individual trajectories. To bridge this gap, we propose Interpretable Knowledge Tracing via Explicit–Implicit Alignment (IKT-EIA), which anchors implicit student knowledge representations with LLM-derived explicit textual insights. IKT-EIA consists of: (1) Context-Guided Knowledge Integration (CGKI) uses LLMs to generate trajectory-aware textual knowledge from student metadata; (2) Query-Driven Knowledge State Alignment (QKSA) treats the implicit state as a query to retrieve and fuse explicit semantics, producing an aligned knowledge state that supports subsequent explanation analysis; (3) Personalized Explainability Adaptation (PEA) generates structured and adaptive feedback by reading the aligned state together with raw interaction records, tailoring explanations to students’ evolving mastery. Experiments on benchmark datasets show that IKT-EIA maintains competitive predictive performance while producing state-aware explanations that were found useful in a controlled evaluation setting. © 2026 Published by Elsevier B.V.