This is the Pytorch implementation for our paper: Efficient Knowledge-aware Recommendation via Distillation.
- torch=1.9.0
- python=3.7.15
- scipy=1.7.3
- numpy=1.21.6
- torch-scatter=2.0.9
We provide three processed datasets and the corresponding knowledge graphs: MovieLens and LastFM and Amazon-book.
You need to create the History/
directory. The command to train EKRD on the CKE teacher model on MovieLens/LastFM/Amazon-book dataset is as follows.
- MovieLens
python main.py --lsreg=3e-4 --aglreg=10 --reg=3e-9 --cdreg=1e-3 --softreg=1e-3 --dataset=movie-lens --teacher_model=CKE
- LastFM
python main.py --lsreg=1e-3 --aglreg=3 --reg=3e-10 --cdreg=1e-3 --softreg=3e-3 --dataset=last-fm --teacher_model=CKE
- Amazon-book
python main.py --lsreg=3e-3 --aglreg=10 --reg=3e-9 --cdreg=3e-3 --softreg=1e-4 --dataset=amazon-book --teacher_model=CKE