MultiScale Spectral GNN for Fraud Detection

This is the PyTorch implementation for the Asonam 2025 paper:

MultiScale Spectral GNN for Fraud Detection
Melike Yildiz Aktas, Mustafa Coskun, Chang-Tien Lu

---

🧠 Framework

---

⚙️ Dependencies

Make sure to install the following requirements:

Python >= 3.8
numpy >= 1.22.4
scipy >= 1.4.1
scikit-learn >= 1.1.2
PyTorch >= 1.11.0
DGL >= 0.9.1

---

## ⚙️ Repo Structure

src/        # All source code scripts
data/       # Original and processed datasets
config/     # Parameter settings (YAML files)
result/     # Model outputs and logs

---

## 🧠 Data

YelpChi.zip	Original YelpChi dataset — hotel and restaurant reviews filtered (spam) or recommended (legit).
Amazon.zip	Amazon Musical Instruments dataset — product reviews.

---

## Configurations

yelp.yaml	Parameters for YelpChi
amazon.yaml	Parameters for Amazon

---

## Model Training

# Unzip the dataset
unzip ./data/YelpChi.zip -d ./data/

# Move to source folder
cd src/

# Convert the dataset to DGL format
python data_preprocess.py --dataset yelp

# Train and test the model
python train.py --dataset yelp