Graph Attention Networks for Organizational Trust Influence Mapping: Identifying Informal Leaders, Bridge Connectors, and Trust Propagation Patterns Using the DEEP Framework

Abstract

Organizational trust does not follow formal hierarchies only. Informal networks via mentorship, collaboration, and peer influence are often the difference between success or failure for a transformation. In this paper, we employ Graph Attention Networks (GATs) to model trust influence patterns in organizations, with the node features being 53-competency trust profiles from the DEEP (Drive, Engage, Empower, Purpose-align) Framework. Trust graphs are constructed from N=12,400 employees in 14 organizations with |E|=47,312 weighted directed edges, and four sources of interaction: formal reporting, 360-degree nominations, project co-participation, and meeting co-attendance. The 2-layer GAT with 4 attention heads achieved a R^2=0.879 on Trust Influence Score (TIS) prediction, which is 46.9% higher than PageRank, 11.1% higher than GCN and 23.5% higher than Node2Vec. The pipeline is formalized by four algorithms: Multi-Source Graph Construction, GAT Trust Influence Prediction, Bridge Leader Identification and Trust Propagation Simulation. Key findings: (1) 23.4% of Exceptional trust leaders lack management titles, suggesting an informal trust infrastructure invisible to org charts; (2) high-trust organizations have 2.3x higher network density and 1.7x higher clustering; (3) ablation analysis shows bridge leaders (4.7% of population) account for 31.2% of organizational trust variance; (4) the Engage dimension is the most effective predictor of cross-cluster influence (β=0.41, p<.001); (5) GAT attention weights show asymmetric trust influence patterns with upward trust flow (subordinate→manager) being 1.4x stronger than downward flow. This work provides a methodology to identify and leverage informal trust infrastructure during purpose-driven organizational transformation.