TCHG: Tri-Trust Conditioned Heterogeneous Graph Learning for Reliable Dynamic Trust Prediction
Researchers have introduced TCHG, a novel framework for dynamic trust prediction that leverages heterogeneous graph learning. Unlike previous methods that treat trust signals uniformly, TCHG decomposes evidence into three distinct channels: entity reliability, interaction-behavior reliability, and contextual trust. Each channel plays a specific role in message propagation and is managed with independent temporal states to ensure accurate predictions, especially in scenarios with sparse or conflicting data. Experiments demonstrate TCHG's effectiveness in improving trust prediction accuracy compared to existing baselines. AI
IMPACT This framework could improve the accuracy of trust prediction systems used in recommendations and fraud detection.