Explainable Educational AI: Integrating Symbolic Reasoning and Diffusion Models for Transparent Knowledge Tracing

ABSTRACT Current educational artificial intelligence (AI) faces challenges in explaining teaching decisions due to insufficient knowledge provenance and transparent reasoning paths. We propose NSRL‐DM‐EDU, a unified framework that integrates neural symbolic reinforcement learning (NSRL) with denoising diffusion probability models (DDPM). Symbolic logic rules are embedded in the policy gradient optimization process of reinforcement learning to ensure that the decision chain adheres to the educational logic, while diffusion modeling reconstructs the trajectory of student knowledge evolution. A path alignment mechanism integrates logical and generative paths to achieve transparent knowledge tracking. Experiments on multidimensional educational data demonstrate that NSRL‐DM‐EDU achieves a causal path consistency score of 0.87, a knowledge node recovery rate of 0.90, and a traceability completeness score of 0.91, outperforming baselines such as (LSTM‐KT), GKT, and BERT‐KT. These results demonstrate that our approach improves explainability and enhances teachers' trust in AI‐driven teaching decisions.