PAA

This is the code for the paper "Political Actor Agent: Simulating Legislative System for Roll Call Votes Prediction with Large Language Models".

Environment

PAA requires API keys for large language models. If you want to use local models (like llama, deepseek), please make sure you have the corresponding models running.

File Structure

agent

The agent folder contains PAA's code:

• agent/decision_pathway.py Uses LLM to analyze committees and caucuses corresponding to different bills
• agent/legislator_agent.py Code for PAA's legislator agents
• agent/prompts.py Prompts used by PAA

data

The data folder contains PAA's data, including experiments mentioned in the paper:

• data/profiles Different profile types used by PAA
• data/district District information
• data/votes Voting data
• data/caucus_committee.py Adds leader agents to legislator profiles
• data/wiki.json Legislator wiki information
• data/caucus_committee.csv Analysis results from agent/decision_pathway.py
• data/caucus_data.json Caucus information
• data/committee_data.json Committee information
• data/caucus_match.json Results of caucus-bill analysis matching
• data/committee_match.json Results of committee-bill analysis matching
• data/name_process.py Used for ablation experiments, replaces legislator names

Running

After processing the profiles files, run main.py in the following format:

python main.py --profiles_path data/profiles/20-profiles433

Citation

If you find our paper helpful, please cite:

@inproceedings{li2025political,
  title={Political actor agent: Simulating legislative system for roll call votes prediction with large language models},
  author={Li, Hao and Gong, Ruoyuan and Jiang, Hao},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={1},
  pages={388--396},
  year={2025}
}