A Time-Aware Self-Supervised Framework for Anomaly Detection in Temporal Graphs

This is the code for A Time-Aware Self-Supervised Framework for Anomaly Detection in Temporal Graphs .

🌐 Project | 📄 Paper

Setup

The code can be run under any environment with Python 3.9.25 and above. (It may run with lower versions, but we have not tested it).

Clone this repo:

git clone https://github.com/slitiWassim/T-ADTG.git
cd T-ADTG/

Install the required packages:

pip install -r requirements.txt

Framework Overview

We evaluate T-ADTG on:

Dataset	Link
Wikipedia
Mooc
Bitcoin-Alpha
Bitcoin-OTC
Amazon

Baselines

Baseline	Paper	Code
Radar
DOMINANT
GDN
SemiGNN
F-FADE
NetWalk
AddGraph
DyRep
TADDY
JODIE
TGAT
TGN
SAD
SLADE
MHisCL

Training

To test T-ADTG on a dataset, run:

 python  main.py \
      --cfg <config-file>  \
      --gpu <device-id> 

For example, to test T-ADTG on Wikipedia:

python main.py \
    --cfg src/config/wikipedia.yaml 

Configuration

• We use YAML for configuration.
• We provide a couple preset configurations.
• Please refer to config.py for documentation on what each configuration does.

Results

Mean ROC curves of T-ADTG on the five benchmark datasets, averaged over 10 runs. Each panel reports the mean AUROC and its standard deviation.

Citing

If you find our work useful, please consider citing:

The paper has been submitted and is currently under review.

Contact

For any question, please file an issue or contact:

Wassim Sliti : wassim.sliti@upm.es

Acknowledgement

This work was carried out within the STRAST Research Group at the Information Processing and Telecommunications Center (IPTC), Universidad Politécnica de Madrid, as part of the CEDAR project, funded by the Horizon Europe Programme (Grant Agreement No. 101135577).