A fairness-aware meta-learning approach for real-time risk offloading decision optimization in digital trade ecosystems

Abstract

This study addresses the critical efficiency-equity trade-off in multi-channel digital trade ecosystems by proposing a fairness-aware meta-learning decision optimization framework for real-time risk offloading under demand shocks. Grounded in the Soft Actor-Critic architecture with composite fairness metrics (Jain index and generalized entropy index) embedded into the meta-optimization objective, the approach employs a two-stage Pareto search strategy to approximate the non-convex efficiency-equity frontier and provide decision-makers with diversified policy options. Experimental validation based on a hybrid discrete-event simulation platform using synthetic transaction and demand-shock data calibrated to the statistical properties of real Wanjiang transaction logs demonstrates that the framework achieves a system throughput of 187.3 req/min and an average Jain fairness index of 0.847, recovering from demand shocks in merely 23.7 time steps—significantly outperforming nine baseline methods. These findings reveal that enforcing explicit fairness constraints (fairness threshold ∈ [0.75, 0.85]) incurs only marginal efficiency penalties (< 3% throughput reduction) while reducing recovery time by 47.6% relative to SAC, offering platform governors a configurable decision support tool to dynamically navigate efficiency-equity trade-offs and foster inclusive digital trade governance.