When Bipartite Graph Learning Meets Anomaly Detection in Attributed Networks: Understand Abnormalities from Each Attribute

This is the official PyTorch implementation of ''When Bipartite Graph Learning Meets Anomaly Detection in Attributed Networks: Understand Abnormalities from Each Attribute'' (Peng, Z, Wang, Y, Lin, Q, Dong, B, & Shen, C, Neural Networks 2025): https://doi.org/10.1016/j.neunet.2025.107194

The code is presented in an easy-to-understand pattern. You could further optimize it based on your own needs.

Requirements

numpy>=1.23.5
scipy>=1.10.0
torch>=1.12.1
dgl>=0.9.0
tensorboard>=2.11.0
scikit-learn>=1.2.1
tqdm>=4.64.1
icecream>=2.1.3
networkx>=2.8.4
matplotlib>=3.7.1
pandas>=1.5.3

Run Enron dataset

We provide Enron dataset for model evaluation.

python main.py --dataset enron --gpu 0

Run your own dataset

1. Process adjacency matrix and feature matrix into scipy.sparse.csr_matrix.
2. Process node labels into numpy dense matrix.
3. Pack these three things above into a dict with key names:

   {'A': adj_matrix, 'X': features, 'gnd': labels}

4. Dump the dict into ./data/mydataset.pickle with pickle module.
5. Run in shell:

python main.py --dataset mydataset --gpu 0

Cite

Please cite our paper if you make advantage of EAGLE in your research:

@article{peng2025bipartite,
  title={When bipartite graph learning meets anomaly detection in attributed networks: Understand abnormalities from each attribute},
  author={Peng, Zhen and Wang, Yunfan and Lin, Qika and Dong, Bo and Shen, Chao},
  journal={Neural Networks},
  pages={107194},
  year={2025},
  publisher={Elsevier}
}