hoangpanda / dragon

DRAGON

Pytorch implementation for "Enhancing Dyadic Relations with Homogeneous Graphs for Multimodal Recommendation" -ECAI'23 arxiv

Data

Data could be download from DropBox: Baby/Sports/Clothing

Run the code

1. Download the data from the data link we provided above, then put the download data to the ./data folder
2. Use conda env create -f dragon.yml to create the enviroment with correct dependencies
3. Run generate-u-u-matrix.py on the dataset you want to generate the user co-occurrence graph
4. Enter the src folder and run with python main.py -m DRAGON -d dataset_name

The parameters to reproduce the result in our paper

Datasets	learning rate	reg weight
Baby	0.0001	0.001
Sports	0.0001	0.001
Clothing	0.0001	0.1

Please consider to cite our paper if this model helps you, thanks:

@article{zhou2023enhancing,
  title={Enhancing Dyadic Relations with Homogeneous Graphs for Multimodal Recommendation},
  author={Zhou, Hongyu and Zhou, Xin and Shen, Zhiqi},
  journal={arXiv preprint arXiv:2301.12097},
  year={2023}
}