StrGNN

The StrGNN repository contains code for an end-to-end structural temporal Graph Neural Network model for detecting anomalous edges in dynamic graphs. The method implemented here is described in this paper.

Installation

Go to detection folder and install required packages

bash install.sh

Dataset Preprocessing

Please follow the example format in data folder to preprocess data:

• acc_email.npy is the graph input with shape (T, N, N), where T is the number of snapshot and N is the number of nodes in the graph

• email0.01 contains train_pos_id, train_neg_id, test_pos_id, test_neg_id, train_pos, train_neg, test_pos, test_neg:

• train_pos_id is with shape (N,). It indicates the which snapshot the edge (in train_pos) come from

• train_pos is with shape (N, 2). Each row i corresponds to a edge in the snapshot train_pos_id[i]

• train_neg_id is with shape (N,). It indicates the which snapshot the edge (in train_neg) comes from

• train_neg is with shape (N, 2). Each row i corresponds to a non-exist edge (sample for training classifier) in the snapshot train_neg_id[i]

• similar for test_pos_id, test_neg_id, test_pos, test_neg

Model Training

Please use example command to train the model. More training setup available in Main.py

python Main.py --graph=acc_email.npy --split=email0.01

Anomaly detection with time-evolving graph. Parameters are the same as accumulate setting, the only difference is that the graph is time-evolving graph

python Main_statistic.py --graph=sta_email.npy --split=email0.01_sta

Citation

If you find the code in this respository useful for your research, please cite our paper:

@inproceedings{cai2021structural,
  title={Structural temporal graph neural networks for anomaly detection in dynamic graphs},
  author={Cai, Lei and Chen, Zhengzhang and Luo, Chen and Gui, Jiaping and Ni, Jingchao and Li, Ding and Chen, Haifeng},
  booktitle={Proceedings of the 30th ACM international conference on Information \& Knowledge Management},
  address = {New York, NY, USA},
  pages={3747--3756},
  year={2021},
  numpages = {10}
}

Acknowledgements

The code is implemented based on

https://github.com/muhanzhang/SEAL

https://github.com/muhanzhang/pytorch_DGCNN

Other information

• Accumulated Graph and Time-evolving Graph:

  Typically, there are two ways to model dynamic systems as dynamic graphs, namely, accumulated graph and time-evolving graph. The accumulated graph automatically and incrementally constructs an entire graph using all vertices and edges until the current timestamp. In contrast, the time-evolving graph is constructed using vertices and edges only appearing in the current timestamp. Therefore, each snapshot represents the graph state at a single instant of time.

License

Creative Commons Attribution-NonCommercial (CC BY-NC) 4.0 International License

The OC4Seq is released under a CC BY-NC 4.0 International License: https://creativecommons.org/licenses/by-nc/4.0.

NonCommercial — You can not use the code for commercial purposes.