Reliability-Oriented Trust Modeling in Open Multi-Agent Networks

Abstract

Reputation mechanisms are widely used to regulate cooperation in decentralized service networks. However, empirical observations suggest that many deployed systems degrade rapidly when coordinated misinformation appears, even in moderate proportions. The vulnerability does not originate solely from malicious agents, but from overly aggressive aggregation strategies. We revisit trust computation from a stability perspective. Instead of computing reputation as a scalar aggregation of direct and indirect feedback, we model interaction histories as weighted temporal graphs and apply damping factors that reflect interaction density and variance. The resulting trust score evolves as a bounded stochastic process rather than a cumulative average. Simulations involving collusive groups and adaptive adversaries show that the proposed formulation reduces oscillation in trust rankings and delays systemic collapse under sustained attack. Rather than eliminating manipulation entirely, the model prioritizes resilience and bounded influence.