TADDY: Anomaly detection in dynamic graphs via transformer

This repo covers an reference implementation for the paper "Anomaly detection in dynamic graphs via transformer" (TADDY).

Some codes are borrowed from Graph-Bert and NetWalk.

Requirments

• Python==3.8
• PyTorch==1.7.1
• Transformers==3.5.1
• Scipy==1.5.2
• Numpy==1.19.2
• Networkx==2.5
• Scikit-learn==0.23.2

Usage

Step 0: Prepare Data

python 0_prepare_data.py --dataset uci

Step 1: Train Model

python 1_train.py --dataset uci --anomaly_per 0.1

Cite

If this code is helpful, please cite the original paper:

@ARTICLE{liu2021anomaly,
  author={Liu, Yixin and Pan, Shirui and Wang, Yu Guang and Xiong, Fei and Wang, Liang and Chen, Qingfeng and Lee, Vincent CS},
  journal={IEEE Transactions on Knowledge and Data Engineering}, 
  title={Anomaly Detection in Dynamic Graphs via Transformer}, 
  year={2021},
  doi={10.1109/TKDE.2021.3124061}}

Don't forget to press a "star" after using!