GloTrust: Bridging local and global views for trust evaluation on blockchain graphs

Trust evaluation serves as a cornerstone for security decision-making in blockchain systems, yet it faces challenges arising from interaction sparsity and the existence of malicious collusions. While Graph Neural Networks (GNNs) show promise, existing local message passing paradigms struggle to capture global trust consensus due to limited receptive fields, and often fail to distinguish the intrinsic asymmetry between trusting behaviors and reputation accumulation. To overcome these limitations, we propose GloTrust, a global modeling framework that bridges local and global views via graph augmentation and pseudo-node routing. Specifically, the framework constructs a d-regular expander graph to introduce long-range connections and employs pseudo nodes as relay routers to establish implicit global communication pathways, thereby enabling global information flow from both structural and routing perspectives. In addition, we design a semantics-aware attention mechanism that adaptively weights different trust-edge types during aggregation, facilitating more effective exploitation of trust semantics. Experiments on two real-world datasets demonstrate that our approach significantly outperforms state-of-the-art baselines across multiple metrics and maintains strong robustness under canonical adversarial scenarios, including good-mouthing and bad-mouthing attacks. Meanwhile, validation results on the Advogato social network dataset demonstrate the consistent superiority and promising generalization potential of GloTrust across different trust domains.