Beyond Redundancy: Information-aware Unsupervised Multiplex Graph Structure Learning

This repository contains the source code and datasets for the NeurIPS'24 paper "Beyond Redundancy: Information-aware Unsupervised Multiplex Graph Structure Learning".

Paper Link: https://openreview.net/pdf?id=xaqPAkJnAS

NeurIPS Project Link (including poster and slides): https://neurips.cc/virtual/2024/poster/93075

The overall framework:

Available Data

All the datasets can be downloaded from datasets link.

Place the 'data' folder from the downloaded files into the 'InfoMGF' directory.

The adjacency matrices of the multiplex graph are extracted from publicly available heterogeneous graph datasets using different meta-paths. By utilizing meta-paths, we can transform the heterogeneous graph into homogeneous graphs with multiplex structures. You can take a look at our code "data_process.py" for the detailed dataset preprocessing process. If you want to learn more about heterogeneous graphs and meta-paths, we recommend the following papers:

1. Wang, Xiao, et al. “Heterogeneous graph attention network.” The World Wide Web Conference. 2019.
2. Shen, Zhixiang, and Zhao Kang. “When Heterophily Meets Heterogeneous Graphs: Latent Graphs Guided Unsupervised Representation Learning.” arXiv preprint arXiv:2409.00687 (2024).

Requirements

This code requires the following:

• Python==3.9.16
• PyTorch==1.13.1
• DGL(cuda)==0.9.1
• Numpy==1.24.2
• Scipy==1.10.1
• Scikit-learn==1.2.1
• Munkres==1.1.4
• kmeans-pytorch==0.3

Training

python main.py -dataset acm

Here, "acm" can be replaced by "dblp", "yelp","mag".

BibTeX

@inproceedings{shen2024beyond,
 author = {Shen, Zhixiang and Wang, Shuo and Kang, Zhao},
 booktitle = {Advances in Neural Information Processing Systems},
 pages = {31629--31658},
 title = {Beyond Redundancy: Information-aware Unsupervised Multiplex Graph Structure Learning},
 volume = {37},
 year = {2024}
}