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.
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
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.