This is the official code repository for the RecSys 2025 paper "Non-parametric Graph Convolution for Re-ranking in Recommendation Systems". It adapts from the public code repository RecBole.

1. Install all the required packages found here.
2. Download all the benchmark datasets used in the paper, including ML-1M, Yelp2018, and Amazon-book. Make sure the folders are named ml-1m, yelp2018, anime and amazon-book respectively.
3. Run the commands in commands.sh for the baseline model training. Save the checkpoint.
4. Run the commands in commands.sh for the baseline model with a graph encoder.
5. Run cal_graph_infos.py to obtain the pkl file for graph_infos. This is the similarity matrix in our strategy. Modify graph_infos_path in recbole_tag.py.
6. Modify and run the commands in commands.sh for applying our strategy to the baseline model.

This repository adapts from the public code repository RecBole.