WhisprRec is a collection of deep learning applications in recommendation algorithms. It mainly collects the mainstream recommendation models and our team's research results in recommendation systems. The framework uses pytorch. IF you find this project is valuable, please star :).
Thanks to the work of Rechorus and Recbole.
- You have to install and configure the following environment: (My environment)
- Python >= 3.8
- Pytorch = 1.9.0
- Pandas = 1.4.4
- Numpy = 1.23.2
- tqdm = 4.51.0
- Clone this repository:
git clone https://github.com/HeyWeCome/WhisprRec.git
- cd into 'src'
cd WhisprRec/src
- Run your or build-in dataset
python main.py --model_name BPRMF --emb_size 64 --lr 1e-3 --l2 1e-6 --dataset ml-100k
In ml-100k dataset.
Model | HR@10 | NDCG@10 | Best hyper-parameters | Description |
---|---|---|---|---|
General | ============ | ============ | ============ | ============ |
BUIR | 0.1114 | 0.0626 | lr=5e-4 | SIGIR'21 |
BPRMF | 0.2254 | 0.1085 | lr=1e-3 | UAI'09 |
LightGCN | 0.2292 | 0.1174 | lr=2e-3, gcn_layers=2 | SIGIR'20 |
SGL | 0.2287 | 0.1187 | lr=2e-3, gcn_layers=2, ssl_tau=0.1, drop_ratio=0.1, type=ED, ssl_weight=0.001 | SIGIR’21 |
Sequential | ============ | ============ | ============ | ============ |
SASRec | 0.3166 | 0.1798 | lr=5e-4, sample=LS, num_layers=1, num_heads=4 | ICDM‘18 |