An autocomplete model for elections

概要

We consider two linked problems for ranked choice elections: building a compact summary of a preference profile $P$, and generating synthetic profiles that resemble $P$ in key ways. We introduce a Markov model based on which candidate names follow which on observed ballots and record this in a succession matrix on candidates together with a start/stop symbol. This provides the basis for new ``autocomplete'' methods for generating synthetic profiles, with ballot length that can be tuned to a target. We compare succession-based generation to parallel use of summaries built from positional frequencies and head-to-head comparisons. To execute this comparison, we introduce new descriptive statistics of profiles that allow new tests of whether synthetic profiles are realistic. In experiments on Scottish ranked-choice profiles, succession-based generation captures held-out ballot structure well, performs especially strongly when completing partial ballots, and often preserves election outcomes, outperforming the alternatives on all counts. Finally, we study the dynamics of iterating the train-and-generate process.

参考リンク

• Paper ↗

リンク先のライセンス・再現性・データ条件をご確認ください。自動分類や実装の目安は保証ではありません。

実装を始めるプロンプト

参考リンクを確認したうえで、次の内容をコピーして利用できます。

# 論文・記事
An autocomplete model for elections

# 概要
We consider two linked problems for ranked choice elections: building a compact summary of a preference profile $P$, and generating synthetic profiles that resemble $P$ in key ways. We introduce a…

# 原典
https://arxiv.org/abs/2610.33035

上記の手法を実装してください。実行前に依存関係とライセンスを確認してください。