guoyang9 / gml-fm

Official implementation for our TKDE paper.

Enhancing Factorization Machines with Generalized Metric Learning

This is the official implementation for our TKDE paper titled with Enhancing Factorization Machines with Generalized Metric Learning. Please follow the flags and directory settings in config.py.

Repo-download

git clone https://github.com/guoyang9/GML-FM.git

Prerequisites

* python==3.7.7
* tqdm==4.31.1
* numpy==1.18.4
* pytorch==1.4.0
* tensorboardX==2.1
* torchvision==0.6.0

Dataset

We use 6 datasets in this paper including three Amazon dataset and one MovieLens dataset and four of them are publicly available.

Pre-processing

1. process the Amazon dataset:

   python pre-process/process_amazon.py
   

2. process the MovieLens dataset:

   python pre-process/process_ml-1m.py
   

Model Training and Auto Evaluation

python main.py --lr 0.001 --batch-size 256 --gpu 0

Citation

If you want to use this code, please cite our paper as follows:

@article{gml-fm,
  title={Enhancing Factorization Machines with Generalized Metric Learning},
  author={Yangyang Guo, Zhiyong Cheng, Jiazheng Jing, Yanpeng Lin, Liqiang Nie, Meng
Wang},
  journal={TKDE}
  year={2020},
  organization={IEEE}
}

gml-fm's Issues

Why isn't feature_valuesused?（model.py:55）

def forward(self, features, feature_values):

Does this function have nothing to do with Feature_values?
evaluate.py:25 predictions = model(features, feature_values)
Does predictions also depend on features only ？