Modeling and optimizing trust-mediated influence in temporal heterogeneous networks.

Public health communication increasingly depends on networked interactions in which attitudes and behavioral intentions emerge through repeated peer exchange and institutional contact, yet trust remains weakly represented in many computational diffusion and graph-learning frameworks, particularly when privacy shocks abruptly reshape credibility. We present a proof-of-concept, mechanism-grounded framework that treats trust as a bounded, directed, and diffusible state on a temporal heterogeneous graph and couples that state to learned influence pathways and budget-constrained intervention optimization. Users, physicians, and privacy-shock events are represented in a timestamped graph; trust changes through event-level directed transfer, relation- and content-dependent modulation, explicitly scoped shock exposure, and parametric recovery. A temporal graph transformer encodes event histories, and influence-aware pooling gates pathway aggregation by source trust before intention prediction and joint node-target and content assignment. Evaluation is conducted entirely in reproducible, mechanism-consistent simulations spanning shock-free, global-shock, and community-targeted-shock regimes. Within these controlled settings, the framework achieved higher predictive performance than the evaluated static, sequential, classical-diffusion, and temporal or heterogeneous baselines, and its optimized policies produced greater simulated intention lift than the evaluated heuristics at the reported operating points. Because the simulator shares structural assumptions with the model, these findings demonstrate internal feasibility rather than established real-world superiority. Empirical calibration, real or semi-real interaction topologies, structurally mismatched policy evaluation, and complete configuration-by-budget uncertainty analyses remain necessary before operational use.