The higher-order networked N-player trust game driven by reputation and reinforcement learning

Trust is a fundamental behavior in human interactions, yet its emergence and maintenance contradict Darwin’s natural selection principle. Existing studies fail to characterize higher-order interactions and overlook the relationship between reputation mechanisms and the environment in Q-learning, both of which hinder the realistic modeling of trust evolution. To address these gaps, this paper proposes a higher-order networked N-player trust game model that integrates reputation mechanisms and reinforcement learning, aiming to investigate how dynamic reputation environments influence trust behaviors. More precisely, we classify agents’ reputations into three levels to form their state set, and define their strategies as the action set. Besides, we introduce reputation correlation and redefine the reward function by integrating agents’ game payoff and reputation state. Simulation results indicate that a higher reputation correlation can effectively promote the evolution of trust and the growth of global wealth, and agent’s reputation evolve toward “Good” dominance. Reinforcement learning parameters (learning rate and discount factor) exert synergistic effects, with larger values of both promoting trust behaviors and global wealth accumulation. We also examine the impact of initial strategy distribution settings, and find that the model exhibits robustness under a high reputation correlation.