Using GAT Structure to Analyze the Propagation Efficiency of Technology Skills Transfer Networks Among Students in Industry-Academia Collaboration

Current studies on technology skills transfer mainly rely on static network models, which have limited capability to characterize the dynamic influence of student heterogeneity on transfer pathways, especially in highly specialized and practice-oriented disciplines. This limitation reduces the accuracy of identifying key skills and efficient propagation paths, thereby constraining precision-oriented industry– academia collaborative training. To address this issue, this paper proposes a Graph Attention Network (GAT)-based framework for analyzing technology skill transfer efficiency. Based on the real learning trajectories of 328 intelligent manufacturing students, a student–skill bipartite network is constructed and projected into a temporally constrained directed transfer graph to preserve the sequential and directional characteristics of skill acquisition. A two-layer GAT with a multi-head attention mechanism is then employed to dynamically learn transfer weights among skills, enabling the identification of asymmetric transfer patterns and heterogeneous learning preferences across different student backgrounds. Experimental results demonstrate a Hit@3 prediction accuracy of 78.4%, while high-propagation skills are primarily concentrated in composite competencies such as system integration and fault diagnosis. The proposed framework establishes an interpretable closed loop integrating dynamic attention modeling, propagation efficiency quantification, and educational intervention, providing measurable and operational support for curriculum optimization and personalized training. Moreover, the graph-based propagation modeling strategy offers methodological insights for intelligent information propagation and graph signal representation in complex engineering systems, with potential relevance to advanced electromagnetic information processing and interdisciplinary learning networks.