This is the source code of PPAC.
Paper: Debiasing Recommendation with Popular Popularity (WWW'24)
conda create -n ppac python=3.8.8
conda activate ppac
pip install torch==1.13.1 --index-url https://download.pytorch.org/whl/cu116
pip install gym==0.23.0 tensorflow-probability==0.20.1 matplotlib scikit-learn
pip install dgl==1.1.3 -f https://data.dgl.ai/wheels/cu116/repo.html
If you want to train PPAC, when you use mf/ncf as base model, please use the following scripts:
python run_MF.py --model ppacmf --dataset {dataset} --train
python run_MF.py --model ppacncf --dataset {dataset} --train
If you want to use LightGCN as base model and train PPAC:
python run_LightGCN.py --model ppaclg --dataset {dataset} --train
${dataset} can be chosen from ['ml-1M', 'gowalla', 'yelp2018'].
For example:
python run_MF.py --model ppacmf --dataset ml-1M --train
After training, if you want to use your pre-trained model to conduct inference, use the below script (remove --train flag).
python run_MF.py --model ppacmf --dataset {dataset} --gamma {gamma} --beta {beta}
python run_MF.py --model ppacncf --dataset {dataset} --gamma {gamma} --beta {beta}
python run_LightGCN.py --model ppaclg --dataset {dataset} --gamma {gamma} --beta {beta}
We provide the hyper-parameters used in our experiments for your references.
| ml-1M | Gowalla | Yelp2018 | ||||
|---|---|---|---|---|---|---|
| BPRMF | 64 | -32 | 512 | -1024 | 256 | -512 |
| NCF | 32 | -16 | 64 | -256 | 128 | -256 |
| LightGCN | 16 | -8 | 64 | -512 | 32 | -128 |
For example:
python run_MF.py --model ppacmf --dataset ml-1M --gamma 64 --beta -32
Since
If you use our datasets or codes, please cite our paper.
@inproceedings{PPAC,
author = {Ning, Wentao and Cheng, Reynold and Yan, Xiao and Kao, Ben and Huo, Nan and Haldar, Nur Al Hasan and Tang, Bo},
title = {Debiasing Recommendation with Popular Popularity},
booktitle = {WWW},
publisher = {ACM},
year = {2024}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent