Overview

HSMAD: Heterophily-Driven Spectral and Manifold Learning for Graph Anomaly Detection

Usage

Command-line arguments

You can run the model by executing the following command:

python main.py --dataset weibo

Arguments

• --dataset (str): Specify the dataset to use. Available options include yelp, weibo, amazon, elliptic, and tolokers.
• --run (int): The number of runs for the experiment. Default is 1.
• --epoch (int): Number of epochs to train the model. Default is 1000.
• --patience (int): Patience for early stopping. The model will stop training if there is no improvement for this many epochs. Default is 100.
• --order (int): The order of filters. Default is 2.
• --q (float): The quantile parameter. Default is 0.5.

Example

To run the model with the weibo dataset, 10 runs, 1000 epochs, patience of 100, execute:

python main.py --dataset weibo --run 10 --epoch 1000 --patience 100 --order 2 --q 0.5

This will train the model using the specified hyperparameters and dataset.

Output

The results of the experiment will include:

• Performance metrics (e.g., Recall, Pecision, F1-Macro, AUROC, AUPRC, G-Mean)