This repo is the official implementation of "Heterogeneous Graph Generation: A Hierarchical Approach using Node Feature Pooling". HG2NP is a heterogeneous graph generation model. The training paradigm is set up as a Generative Adversarial Network (GAN).
Model - Refers to HG2NP (Our) Implementation
DiGress - Used to create skeleton homogeneous graphs.
HGEN - Another heterogeneous graph generation that was tried out but couldn't be fully adapted.
MAGNN - Repository using which the IMDB and DBLP datasets were preprocessed.
NetGAN, VGAE - Other graph generation methods that were used for comparisons.
If you use HG2NP for research purposes, please cite our paper:
@article{ghosh2024heterogeneousgraphgenerationhierarchical,
title={Heterogeneous Graph Generation: A Hierarchical Approach using Node Feature Pooling},
author={Hritaban Ghosh and Chen Changyu and Arunesh Sinha and Shamik Sural},
journal={arXiv preprint arXiv:2502.05564},
year={2024}
}