hardenkim / torch-recommenders

torch-recommenders

Recommendation System Models by Pytorch

Models

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.

Model Comparison

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.

MovieLens-1m

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

KMRD-2m

Model	mAP@k	nDCG@k	HR@k
SASRec	0.757	0.805	0.955

Examples

• 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

Requirements

* 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)

References

• rixwew/pytorch-fm
• pmixer/SASRec.pytorch
• pyy0715/Neural-Collaborative-Filtering