Resilient Trust Evaluation Framework for Large-Scale Multi-Agent Systems

Abstract

In open multi-agent environments, cooperation among autonomous entities depends heavily on reliable trust evaluation mechanisms. However, reputation-based systems remain vulnerable to collusion, misinformation dissemination, and strategic manipulation. This paper proposes a resilience-oriented trust evaluation framework that integrates behavioral modeling, network topology analysis, and adversarial simulation. Agent interaction patterns are represented as dynamic trust graphs, and probabilistic risk indicators are derived through multi-stage aggregation. A simulation platform is developed to evaluate system robustness under diverse attack strategies. Experimental studies demonstrate that the proposed framework can effectively identify structural vulnerabilities and improve long-term cooperation stability.