Reverse-Engineering Fan Votes in Dancing with the Stars: Rule-Consistent Inference, Fairness Analysis, and Mechanism Design
Journal of Computer Science and Artificial Intelligence · 28.07.2026 · 8 (1) · 42-54
DOI: 10.54097/5yy1d380 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Computer Science and Artificial Intelligence, online erschienen am 28.07.2026. Verfasst von Yichen Wang und 2 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
This paper develops a data-driven framework to infer confidential audience voting and to evaluate voting aggregation mechanisms in Dancing with the Stars under rule changes and structural bias. We propose a two-stage inverse approach. In Stage 1, a Bradley Terry style model is fitted to historical elimination sequences to estimate each contestants latent fan strength, forming a prior preference distribution. In Stage 2, we reconstruct weekly audience vote shares by solving a constrained inverse problem: among all vote allocations consistent with the show’ s elimination mechanism (including percentage-based elimination and the bottom-two variant), we select the distribution closest to the prior via convex quadratic programming (with discrete handling for bottom-two weeks). Uncertainty is quantified through posterior sampling. Across 34 seasons, the reconstructed votes reproduce 77.0% of weekly eliminations, achieve a 0.953 Spearman correlation with final rankings, and match champions 58.8% of the time, indicating strong explanatory power despite unobserved votes. Using the inferred votes, we compare historical aggregation rules and find the percentage-based rule aligns substantially better with observed eliminations than the rank-based rule in their respective eras (78.6% vs. 56.0 weekly agreement), while avoiding excessive amplification of judge influence. We further quantify structural fan advantage related to celebrity background and propose a debiasing weight to correct audience shares. Building on this correction, we introduce a Fairness-Adjusted Hybrid Rule (FHR) that blends judge shares with debiased audience shares. FHR is designed as a forward-looking mechanism: it improves end-of-season technical ordering (finals rank correlation 0.325 vs. 0.271 under the base- line percentage rule) at the cost of lower week-by-week elimination consistency, and its debiasing strength can be tuned to match production priorities. We recommend retaining the percentage-based rule as the default for stability and historical continuity, while piloting the FHR as an optional fairness overlay with transparent parameter tuning and periodic audits.
Zitieren
APA
Wang, Y., Chen, Z., & Liu, Q. (2026). Reverse-Engineering Fan Votes in Dancing with the Stars: Rule-Consistent Inference, Fairness Analysis, and Mechanism Design. Journal of Computer Science and Artificial Intelligence, 8(1), 42-54. https://doi.org/10.54097/5yy1d380
Vancouver
Wang Y, Chen Z, Liu Q. Reverse-Engineering Fan Votes in Dancing with the Stars: Rule-Consistent Inference, Fairness Analysis, and Mechanism Design. Journal of Computer Science and Artificial Intelligence. 2026;8(1):42-54. doi:10.54097/5yy1d380
BibTeX
@article{Wang2026,
author = {Wang, Yichen and Chen, Zhan and Liu, Qian},
title = {Reverse-Engineering Fan Votes in Dancing with the Stars: Rule-Consistent Inference, Fairness Analysis, and Mechanism Design},
journal = {Journal of Computer Science and Artificial Intelligence},
year = {2026},
volume = {8},
number = {1},
pages = {42--54},
doi = {10.54097/5yy1d380}
}
EUVIMED