README: Inductive Anomaly Detection in Dynamic Graphs

Overview

This repository provides an implementation of the Dynamic Graph Transformer for Inductive Anomaly Detection, incorporating the Accumulative Causal Walk Alignment (ACWA) mechanism.

Data

To reproduce the experiments, please ensure you have the corresponding dataset files (Bitcoin-Alpha, Bitcoin-OTC, Digg, UC-Ivrine Messages) available in the data/ directory.

Installation

Requirements:

• Python 3.8+
• PyTorch 1.9+
• NetworkX
• NumPy
• Matplotlib
• Scikit-learn

To install the dependencies, run:

pip install -r requirements.txt

Running Experiments

Precompute Node Embeddings

To precompute CTDNE-based node embeddings, use the scripts in:

python acwa.py

Train and Evaluate the Model

To train the Dynamic Graph Transformer for anomaly detection, use the scripts in:

python dcidgt_anomalydet_main.py