Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval

This repository contains the code for the paper "Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval," which has been accepted by ICKG2024. The project aims to enhance the capabilities of role-playing agents through emotional retrieval.

Environment Requirements

• Python Version: 3.12
• Other Library Versions: See requirements.txt

Installation Steps

STEP 1: Prepare Models and Configuration

1. Download Base Models:

   • Qwen/Qwen2-7B-Instruct
   • BAAI/bge-base-zh-v1.5

2. Configure Model Paths and GPT Keys:

   • Configure local model paths and GPT keys in utils/config.py.
   • Configure GPT keys for evaluation in evaluation/config.json.

STEP 2: Prepare Datasets

1. Download CharacterEval Dataset:

   wget https://github.com/morecry/CharacterEval/raw/refs/heads/main/data/test_data.jsonl
   wget https://raw.githubusercontent.com/morecry/CharacterEval/refs/heads/main/data/character_profiles.json

   Place the files from the CharacterEval dataset into the data/charactereval directory.

2. Generate query_bank:

   python data/generate_query_bank.py

3. Generate the All Memory Bank for All Characters:

   python data/charactereval/generate_qa_memory_bank.py

STEP 3: Generate Psychology Questionnaire Answers and Evaluate

1. Generate Psychology Questionnaire Answers:

   python get_response.py

2. Conduct Evaluation:

   python evaluation/run_experiments.py

   The evaluation plan uses the personality assessment from InCharacter.

Citation

If you use this project in your research, please cite the following paper:

@misc{huang2024emotionalragenhancingroleplaying,
      title={Emotional RAG: Enhancing Role-Playing Agents through Emotional Retrieval}, 
      author={Le Huang and Hengzhi Lan and Zijun Sun and Chuan Shi and Ting Bai},
      year={2024},
      eprint={2410.23041},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2410.23041},  
}