EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Bayesian Inverse Sampling and Shapley-Driven Calibration for Dual-Channel Preference Aggregation

Jiawei Liu, Kai Li, Qisheng Su

Academic Journal of Applied Sciences · 13.07.2026 · 2 (2) · 71-76

DOI: 10.54097/erpkp016 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Academic Journal of Applied Sciences, online erschienen am 13.07.2026. Verfasst von Jiawei Liu und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper presents a dual-channel preference aggregation framework that reconstructs latent user voting distributions and calibrates heterogeneous scoring channels through Bayesian inverse sampling and gradient boosting attribution. The target setting is a mixed-evaluation ranking system where expert scores and user votes drive elimination jointly. Three coupled problems are addressed: latent preference recovery from elimination outcomes, normalization-induced bias quantification across scoring schemes, and dynamic weight allocation under temporal drift. A Monte Carlo inverse sampling module reconstructs latent voting distributions across 34 episodes by enforcing elimination-consistency constraints with Bayesian priors, raising the simulation acceptance rate from 28.6% to 44.6%. A share dispersion metric quantifies variance convergence near elimination boundaries, with eliminated candidates registering dispersion 0.0252 against 0.0601 for advancing candidates. The Fan Dominance Index measures normalization-induced bias. Percentage-based aggregation amplifies user signals more aggressively than rank-based aggregation, while a normalized expert override neutralizes excessive popularity influence. An XGBoost regression with Shapley attribution reaches R-squared of 0.906, with candidate age (weight 0.324) as the dominant judging predictor and a per-expert efficacy hierarchy that registers marginal premiums up to +0.603. The D.A.R.E. system integrates dual-quantile calibration and dynamic weighting, and remains robust under 275 weeks of simulation with extreme voting fluctuations.

Zitieren

APA

Liu, J., Li, K., & Su, Q. (2026). Bayesian Inverse Sampling and Shapley-Driven Calibration for Dual-Channel Preference Aggregation. Academic Journal of Applied Sciences, 2(2), 71-76. https://doi.org/10.54097/erpkp016

Vancouver

Liu J, Li K, Su Q. Bayesian Inverse Sampling and Shapley-Driven Calibration for Dual-Channel Preference Aggregation. Academic Journal of Applied Sciences. 2026;2(2):71-76. doi:10.54097/erpkp016

BibTeX
@article{Liu2026,
  author = {Liu, Jiawei and Li, Kai and Su, Qisheng},
  title = {Bayesian Inverse Sampling and Shapley-Driven Calibration for Dual-Channel Preference Aggregation},
  journal = {Academic Journal of Applied Sciences},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {71--76},
  doi = {10.54097/erpkp016}
}
Kostenlos starten →