A dynamic multiobjective optimization framework for hybrid evaluation systems: a Bayesian inverse and adaptive weighting approach

Abstract

Hybrid scoring systems integrating professional technical assessment and public preference often face critical challenges in balancing competitive fairness and public engagement. This paper proposes a multi-stage mathematical framework to optimize such dual-scoring ecosystems. First, unobserved public preference behavior is reconstructed via a Bayesian inverse approach with Monte Carlo simulations over historical data to estimate latent distributions. Second, counterfactual historical replay is performed to evaluate alternative scoring mechanisms, showing that static proportional rules tend to amplify polarized public opinions. Key performance drivers are then disentangled using a Linear Mixed-Effects Model (LMM) to isolate heterogeneous contribution effects. Finally, an Adaptive Fair-Engage System (AFES) is developed, leveraging dynamic weighting and controversy-aware penalty mechanisms to achieve a Pareto-optimal trade-off between technical integrity and engagement suspense. Backtest results validate that AFES outperforms traditional static methods in fairness while maintaining higher engagement.