The Performance of All Models on the Gowalla Dataset
| Rank | Model | Recall@20 | NDCG@20 | Paper | Year |
|---|---|---|---|---|---|
| 1 | NESCL | 0.1917 | 0.1617 | Neighborhood-Enhanced Supervised Contrastive Learning for Collaborative Filtering | 2024 |
| 2 | BSPM-EM | 0.192 | 0.1597 | Blurring-Sharpening Process Models for Collaborative Filtering | 2022 |
| 3 | BSPM-LM | 0.1901 | 0.157 | Blurring-Sharpening Process Models for Collaborative Filtering | 2022 |
| 4 | LT-OCF | 0.1875 | 0.1574 | LT-OCF: Learnable-Time ODE-based Collaborative Filtering | 2021 |
| 5 | SimpleX | 0.1872 | 0.1557 | SimpleX: A Simple and Strong Baseline for Collaborative Filtering | 2021 |
| 6 | UltraGCN | 0.1862 | 0.158 | UltraGCN: Ultra Simplification of Graph Convolutional Networks for Recommendation | 2021 |
| 7 | Emb-GCN | 0.1862 | UltraGCN: Ultra Simplification of Graph Convolutional Networks for Recommendation | 2021 | |
| 8 | GF-CF | 0.1849 | 0.1518 | How Powerful is Graph Convolution for Recommendation? | 2021 |
| 9 | LightGCN | 0.183 | 0.1554 | LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation | 2020 |
| 10 | NGCF | 0.157 | Neural Graph Collaborative Filtering | 2019 |
-
The code repository for the paper: Peijie Sun , Le Wu, Kun Zhang, Xiangzhi Chen, Meng Wang. Neighborhood-Enhanced Supervised Contrastive Learning for Collaborative Filtering (Accepted by TKDE).
-
The dataset can refer to following links(Baidu Netdisk, Google Drive).
-
The parameters files locate in config/amazon-book \ gowalla \ yelp2018 directories
-
As we have updated the proposed model name to NESCL, its previous name is SUPCCL, it can be found in the path recbole/model/general_recommender/supccl.py
-
To train the model, you should first prepare the training environment
pip install -r requirements.txtpython setup.py build_ext --inplace(We adopt the C++ evaluator in https://github.com/kuandeng/LightGCN)
-
Then, you can execute following commands to train the model based on different datasets:
-
python run_recbole_autodl.py --model=SUPCCL --dataset=yelp2018 --config=True --dataloader_file=/root/autodl-fs/yelp2018-for-SUPCCL-dataloader.pth -
python run_recbole_autodl.py --model=SUPCCL --dataset=amazon-book --config=True --dataloader_file=/root/autodl-fs/amazon-book-for-SUPCCL-dataloader.pth -
python run_recbole_autodl.py --model=SUPCCL --dataset=gowalla --config=True --dataloader_file=/root/autodl-fs/gowalla-for-SUPCCL-dataloader.pth
- The generated log files saved in
logdirectory, and the temporal model parameters can saved in thesaveddirectory.
If you are interested in my work, you can also pay attention to my personal website: https://www.peijiesun.com
You can cite our paper with:
@article{sun2023neighborhood,
title={Neighborhood-Enhanced Supervised Contrastive Learning for Collaborative Filtering},
author={Sun, Peijie and Wu, Le and Zhang, Kun and Chen, Xiangzhi and Wang, Meng},
journal={IEEE Transactions on Knowledge and Data Engineering},
year={2023},
publisher={IEEE}
}