This repo contains the codes of the simulation platform equipped with CreAgent, which is used for long-term recommender system evaluation. proposed by the SIGIR 2025 paper "LLM-Empowered Creator Simulation for Long-Term Evaluation of Recommender Systems Under Information Asymmetry"
Clone the github repo and create the conda environment
git clone https://github.com/shawnye2000/CreAgent.git
# Create conda environment
conda create -n creagent python=3.10
conda activate creagent
# Install requirements
cd CreAgent
pip install -r requirements.txt
Then, download the dataset from google drive. Download the users.json and provider.json from Small_YouTube and put them into dataset/youtube.
Third, please determine the setups of the simulation platform and modify the config file:config/config.yaml
api_base: http://localhost:8000/v1
api_key: EMPTY
llm_model_name: your_llm_name
embedding_model_path: your_embedding_model
Activate your conda environment
conda activate creagent
Load your vllm
python -m vllm.entrypoints.openai.api_server --model your_llm_name --trust-remote-code --tensor-parallel-size 2 --api-key EMPTY --port 8000 --enforce-eager --gpu-memory-utilization 0.9
Then, you can run the simulator environment
python simulator/simulator.py
To change the configure setting, you can enter the config/config.yaml file and edit.