Recommendation System Models by Pytorch
Model | Paper |
---|---|
Factorization Machine | S Rendle, "Factorization Machines", 2010. |
Field-aware Factorization Machines | Y Juan, et al. "Field-aware Factorization Machines for CTR Prediction", 2015. |
Wide&Deep | HT Cheng, et al. "Wide & Deep Learning for Recommender Systems", 2016. |
DeepFM | H Guo, et al. "DeepFM: A Factorization-Machine based Neural Network for CTR Prediction", 2017. |
Graph Convolutional Networks | Kipf & Welling. "Semi-Supervised Classification with Graph Convolutional Networks", 2016. |
Neural Collaborative Filtering | He, Xiangnan, et al. "Neural collaborative filtering", 2017. |
Self-Attentive Sequential Recommendation | W. Kang and J. McAuley, "Self-Attentive Sequential Recommendation", 2018. |
MovieLens(ml-1m) is used as the dataset for model comparision.
To evaluate the performance of item recommendation, I adopted the leave-one-out
evaluation, which has been widely used in many literatures (NCF, Wide&Deep, SASRec).
Use k=10 (top 10 recommendations) for ranking metrics.
The Hyperparameters for each model are in model.ini
.
Model | mAP@k | nDCG@k | HR@k |
---|---|---|---|
FM | 0.353 | 0.439 | 0.716 |
FFM | 0.355 | 0.441 | 0.717 |
Wide&Deep | 0.306 | 0.389 | 0.659 |
DeepFM | 0.341 | 0.424 | 0.693 |
NeuMF | 0.339 | 0.422 | 0.692 |
SASRec | 0.478 | 0.554 | 0.797 |
Model | mAP@k | nDCG@k | HR@k |
---|---|---|---|
SASRec | 0.757 | 0.805 | 0.955 |
- FM
python main.py --model=fm
- FFM
python main.py --model=ffm
- Wide & Deep
python main.py --model=wd
- DeepFM
python main.py --model=dfm
- NMF
python main.py --model=nmf
- SASRec
python main_sasrec.py --model=sasrec
* python>=3.8.12
* pytorch>=1.10.2
* numpy>=1.21.2
* pandas>=1.3.5
* scipy>=1.5.4
* tensorboardX>=2.5 (mainly useful when you want to visulize the loss, see https://github.com/lanpa/tensorboard-pytorch)