FIAD: Graph anomaly detection framework based feature injection

You can check the paper here [pdf].

Environment Setup

Before you start, install Pytorch and torch-geometric with appropriate CUDA support. Please refer to the PyTorch and torch-geometric websites for the specific guidelines.

My environment is as follows:

torch==1.13.0
torchaudio==0.13.0
torchvision==0.14.0
torch-cluster==1.6.1
torch-scatter==2.1.1
torch-sparse==0.6.17
torch-spline-conv==1.2.2
torch-geometric==2.3.1

Install other dependencies:

pip install -r requirements.txt

Dataset

• books, weibo, and reddit datasets. please refer to here.

• cora, citeseer, and pubmed datasets. please refer to here.

• ogbn-arxiv dataset. please refer to here.

Experiments

Parameters:

• dataset: dataset name.
• dataset_dir: the directory where the dataset is located.
• batch_size: batch size of graphs. 0 for total graph.

Example:

python main.py --dataset weibo --dataset_dir ./data/ --batch_size 0

You can further modify the parameters in the paper within the main.py file.

hid_dim = [...]         # Embedding dimension: ℎ
lr = [...]              # Learning Rate
injection_rate = [...]  # Proportion of injection: 𝑝
alpha = [...]           # Proportion between attribute and structure: 𝛼
beta = [...]            # Proportion between two losses: 𝛽

The log folder contains a portion of the training model records.

Citing FIAD:

@article{FIAD2025CHEN,
	title = {FIAD: Graph Anomaly Detection Framework Based Feature Injection},
	author = {Chen, Aoge and Wu, Jianshe and Zhang, Hongtao},
	year = {2025},
	journal = {Expert Systems with Applications},
	volume = {259},
	pages = {125216},
	doi = {10.1016/j.eswa.2024.125216}
}