README.md

LLMInit: Selective Initialization from Large Language Models for Recommendation

This is the PyTorch Implementation for LLMInit.

Environment Setup

1. Create a new conda environment:

conda create -n llminit python=3.8
conda activate llminit

2. Install PyTorch (adjust CUDA version as needed):

conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia

3. Install RecBole and other dependencies:

pip install recbole
pip install transformers
pip install scikit-learn

Data Processing

The data processing is automatically handled by RecBole framework. When you run the code with our provided configuration:

1. The dataset (e.g., Amazon-Beauty) will be automatically downloaded
2. Data preprocessing will be performed automatically
3. The processed data will be cached for future use

No manual data processing is required.

Usage Examples

(1) run the LLMInit-Rand with the LightGCN on Amazon-Beauty

python run_recbole.py --opt rand -d amazon-beauty -m ContGCN

(2) run the LLMInit-Uni with the SGL on Amazon-Beauty

python run_recbole.py --opt uni -d amazon-beauty -m ContSGL

(3) run the LLMInit-Var with the SGCL on Amazon-Beauty

python run_recbole.py --opt var -d amazon-beauty -m ContSGCL

Acknowledgement

The structure of this repo is built based on RecBole. Thanks for their great work.