Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Persona-driven Simulation of Voting Behavior in the European Parliament with Large Language Models

Created by
  • Haebom

Author

Maximilian Kreutner, Marlene Lutz, Markus Strohmaier

Outline

This paper explores predicting the voting behavior of Members of the European Parliament (MEPs) by leveraging the political bias of large-scale language models (LLMs). Given the LLMs' tendency toward a left-wing liberal orientation, we used a zero-shot persona prompting technique with limited information to predict individual MEPs' voting decisions and the policy positions of European groups. We evaluated the robustness of the predictions using various persona prompts and generation methods, and found that the model simulates the voting behavior of MEPs reasonably well, with a weighted F1 score of approximately 0.793. The politician persona dataset and code used are publicly available.

Takeaways, Limitations

•
Takeaways:
◦
We show that LLM can be used to predict political voting behavior with considerable accuracy even with limited information.
◦
We demonstrate that the political bias of LLM can be exploited to simulate diverse political positions through the persona prompting technique.
◦
Presenting a new methodology that can help understand the decision-making process in complex political systems such as the European Parliament.
◦
Expanding research and application possibilities through open datasets and code.
•
Limitations:
◦
LLM's high reliance on political bias limits its credibility unless the bias itself is addressed.
◦
The performance of zero-shot prompting can be highly dependent on prompt engineering and data quality.
◦
The prediction accuracy is satisfactory with a weighted F1 score of 0.793, but it is not perfect, and further analysis of the prediction error range is required.
◦
Because the study was limited to the European Parliament, caution should be exercised in generalizing to other political systems.
👍
Made with Slashpage