Network-Aware Distributed Learning Path Optimization via Graph-Based Knowledge Tracing

High-adaptability learning path optimization aims to generate personalized learning trajectories that align with learners' cognitive characteristics based on their current knowledge states and learning goals, thereby enhancing learning efficiency. Despite progress in cognitive modeling and path recommendation, existing methods face three major challenges. First, limited structural modeling capacity hinders accurate representation of hierarchical relations and dependencies among knowledge concepts. Second, most methods overlook the dynamic evolution of learners' cognitive states, resulting in insufficient adaptability and personalization. Third, they typically assume local resource availability, failing to account for network constraints in distributed environments, which leads to poor accessibility and high latency in real-world deployments. To address these challenges, we propose a network-aware distributed learning path optimization approach based on graph-based knowledge tracing. At the modeling level, we integrate subject knowledge graphs, learners' knowledge states, and resource distribution states to construct a state space that supports both cognitive dynamics and network adaptivity. At the optimization level, we formulate the path planning task as a sequential decision-making process in reinforcement learning and design a novel PN-D3QN algorithm, which integrates Prioritized Experience Replay (PER) and Noisy Neural Network (NoisyNet) into a Dueling Double Deep Q-Network to jointly optimize cognitive relevance and execution feasibility. Experimental results on multiple benchmark datasets demonstrate the superiority of our method over existing solutions, significantly improving both learning effectiveness and system adaptability.