/BoxGNN

Primary LanguagePython

BoxGNN

This is our PyTorch implementation for the paper:

Fake Lin, Ziwei Zhao, Xi Zhu, Da Zhang, Shitian Shen, Xueying Li, Tong Xu, Suojuan Zhang and Enhong Chen (2024). When Box Meets Graph Neural Network in Tag-aware Recommendation. In KDD ’24, August 25–29, 2024, Barcelona, Spain.

Reproducibility

MovieLens

python main.py --dataset movielens --model boxgnn --beta 0.2

LastFm

python main.py --dataset lastfm --model boxgnn --beta 0.3

E-shop

python main.py --dataset e-shop --model boxgnn --beta 0.2 --context_hops 2

Preprocess

The preprocess details are listed in utils/preprocess.py, and the raw data of movielens and lastfm come from hetrec-2011. Notably, we randomly select 80%, 10%, 10% of data as training set, validation set and test set, respectively. We only use data in training set to construct collaborative tag graph.