Inverse Reconstruction of Latent Voting and Rule Impact Evaluation Based on Genetic Algorithm and Monte Carlo Counterfactual Simulation

Abstract

This paper addresses the issue of unobservable fan voting in competitive reality shows by constructing a comprehensive modeling framework based on reverse inference and random simulation. For scenarios where voting proportions are latent variables and elimination mechanisms exhibit nonlinear ranking characteristics, the voting reconstruction is formalized as a constrained high-dimensional optimization problem. A genetic algorithm is employed for global search to estimate the full-season voting distribution under normalization conditions, enabling robust inversion of the implicit voting structure. Building upon this foundation, a Logistic probability model integrating judge scores and estimated voting characteristics is constructed to predict and validate elimination outcomes. Bootstrap intervals and cross-validation are employed to enhance model stability. Furthermore, a Monte Carlo-based counterfactual simulation framework is proposed to propagate voting uncertainty across different scoring rules. Rule deviations and system sensitivity are quantified using fan influence indices and outcome variance rates. A dynamic simulation model incorporating multi-week state transitions and intervention mechanisms is constructed to analyze the long-term impact of rule adjustments on competitive trajectories. This methodology achieves a unified modeling approach for latent variable reconstruction, probabilistic prediction, and institutional evaluation, demonstrating strong universality and scalability.