From Simulation to Strategy: Automating Personalized Interaction Planning for Conversational Agents

Environment

• Python 3.11
• Install dependencies from requirements.txt

Persona Generation

Run the following scripts to generate personas:

Note: You will need the Meta-Llama-3.1-8B-Instruct-Q8_0.gguf model file for persona generation.

python generate_persona_age_gender.py
python generate_persona_occupation.py

Simulation

• For Persona Attribute Analysis, run simulation_exp1.sh.
  Ensure you are hosting Llama-3.1-8B-Instruct and update the endpoint URL in the code as needed.

• For Occupation-Based Strategy for SalesAgent, run simulation_exp2_no_strategy.sh and simulation_exp2_strategy.sh.
  You will need to host Qwen3-8B and mistral-7b-instruct-v0.3, and modify the endpoint URLs accordingly.

Analysis

Refer to the following notebooks for analysis:

• age_gender_analysis.ipynb
• occupation_analysis.ipynb

Citation

If you find this paper useful in your work, please cite our paper.

@misc{chang2025simulationstrategyautomatingpersonalized,
      title={From Simulation to Strategy: Automating Personalized Interaction Planning for Conversational Agents},
      author={Wen-Yu Chang and Tzu-Hung Huang and Chih-Ho Chen and Yun-Nung Chen},
      year={2025},
      eprint={2510.08621},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.08621},
}