Heuristic Surveying Trust in Networks: Classical Models and Modern Deep Learning Approaches

Abstract

Early trust frameworks relied on heuristic or graph-theoretic rules (e.g., EigenTrust, TrustRank) but often suffered from brittle propagation semantics and limited adaptability to dynamic behaviors. In recent years, data-driven approaches—particularly graph neural networks (GNNs), attention-based models, and hybrid deep learning frameworks—have catalyzed significant advances in trust evaluation by learning the latent representation of trust features and propagation patterns directly from data. This review presents a comprehensive survey of network trust research, encompassing foundational definition and taxonomies, classical computational methods, machine learning and deep learning innovations (with emphasis on GNNs and attention mechanisms), evaluation protocols and datasets, key applications (social recommender systems, IoT security, blockchain, and fraud detection), as well as ongoing challenges (data sparsity dynamic adaptation, explainability, and robustness). Network trust quantifies the confidence in relationships among entities (e.g., users, devices, organizations) in interconnected systems, ranging from social networks to the Internet of Things (IoT) and peer-to-peer (P2P) environments. We conclude by outlining promising research directions, such as self-supervised trust representation learning, lifelong adaptation, explainable trust frameworks, privacy-preserving trust computation, and multimodal trust fusion, to guide future work toward robust, scalable, and interpretable trust mechanisms. This review synthesizes the current state of the art in network trust, providing researchers and practitioners with a structured roadmap of methodologies, datasets, applications, and future research directions.