Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ElectionSim: Massive Population Election Simulation Powered by Large Language Model Driven Agents

Xinnong Zhang, Jiayu Lin, Libo Sun, Weihong Qi, Yihang Yang, Yue Chen, Hanjia Lyu, Xinyi Mou, Siming Chen, Jiebo Luo, Xuanjing Huang, Shiping Tang, Zhongyu Wei

TL;DR

ElectionSim presents a scalable framework for massive population election simulation using LLM-driven agents and a million-plus voter pool derived from social media. It tackles demographic alignment with real-world statistics via IPF-based distribution sampling and introduces PPE, a poll-based benchmark for rigorous evaluation at voter and state levels. Results show strong voter-wise accuracy (Micro-F1 >80% on voting-related items) and state-level prediction success across most states, supported by both prompt-based baselines and ablation analyses. The work also provides an interactive visualization platform to explore macro and micro election dynamics, demonstrating robustness and practical applicability for U.S. presidential election scenarios.

Abstract

The massive population election simulation aims to model the preferences of specific groups in particular election scenarios. It has garnered significant attention for its potential to forecast real-world social trends. Traditional agent-based modeling (ABM) methods are constrained by their ability to incorporate complex individual background information and provide interactive prediction results. In this paper, we introduce ElectionSim, an innovative election simulation framework based on large language models, designed to support accurate voter simulations and customized distributions, together with an interactive platform to dialogue with simulated voters. We present a million-level voter pool sampled from social media platforms to support accurate individual simulation. We also introduce PPE, a poll-based presidential election benchmark to assess the performance of our framework under the U.S. presidential election scenario. Through extensive experiments and analyses, we demonstrate the effectiveness and robustness of our framework in U.S. presidential election simulations.

ElectionSim: Massive Population Election Simulation Powered by Large Language Model Driven Agents

TL;DR

ElectionSim presents a scalable framework for massive population election simulation using LLM-driven agents and a million-plus voter pool derived from social media. It tackles demographic alignment with real-world statistics via IPF-based distribution sampling and introduces PPE, a poll-based benchmark for rigorous evaluation at voter and state levels. Results show strong voter-wise accuracy (Micro-F1 >80% on voting-related items) and state-level prediction success across most states, supported by both prompt-based baselines and ablation analyses. The work also provides an interactive visualization platform to explore macro and micro election dynamics, demonstrating robustness and practical applicability for U.S. presidential election scenarios.

Abstract

The massive population election simulation aims to model the preferences of specific groups in particular election scenarios. It has garnered significant attention for its potential to forecast real-world social trends. Traditional agent-based modeling (ABM) methods are constrained by their ability to incorporate complex individual background information and provide interactive prediction results. In this paper, we introduce ElectionSim, an innovative election simulation framework based on large language models, designed to support accurate voter simulations and customized distributions, together with an interactive platform to dialogue with simulated voters. We present a million-level voter pool sampled from social media platforms to support accurate individual simulation. We also introduce PPE, a poll-based presidential election benchmark to assess the performance of our framework under the U.S. presidential election scenario. Through extensive experiments and analyses, we demonstrate the effectiveness and robustness of our framework in U.S. presidential election simulations.

Paper Structure

This paper contains 73 sections, 3 equations, 14 figures, 14 tables.

Table of Contents

  1. Introduction
  2. ElectionSim: Massive Population Election Simulation
  3. Massive and Diverse Voter Pool Construction
  4. Social Media Data Collection
  5. Source Data
  6. Data Preprocessing
  7. Demographic Feature Annotation
  8. Demographic Taxonomy
  9. Demographic Classifier
  10. Real-world Demographic Distribution Sampling Strategy
  11. Real-World Demographics in Electoral Processes
  12. Methods for Distribution Sampling
  13. Iterative Proportional Fitting
  14. PPE: Poll-based Presidential Election Benchmark
  15. Task Definition
  16. ...and 58 more sections

Figures (14)

  • Figure 1: Simulation results of the 2020 Presidential Election. The colors represent the real-world results and the marks represent the simulation results accuracy.
  • Figure 2: An illustration of the overall pipeline of the U.S. presidential election simulations.
  • Figure 3: Statistical charts based on user aggregation, including: a) Density plot of tweets per user, b) Density plot of average word count per tweet, c) User distribution across languages, and d) Density plot of user overlap scores.
  • Figure 4: Classification accuracy of API-based LLMs and our demographic classifiers.
  • Figure 5: Distribution of attribute categories in the voter pool
  • ...and 9 more figures