Skip to content

Latest commit

 

History

History
192 lines (159 loc) · 9.64 KB

README.md

File metadata and controls

192 lines (159 loc) · 9.64 KB
Raw



PRs Welcome GitHub GitHub release PRs

A Deep Graph-based Toolbox for Fraud Detection

Introduction

May 2021 Update: The DGFraud has upgraded to TensorFlow 2.0! Please check out DGFraud-TF2

DGFraud is a Graph Neural Network (GNN) based toolbox for fraud detection. It integrates the implementation & comparison of state-of-the-art GNN-based fraud detection models. The introduction of implemented models can be found here.

We welcome contributions on adding new fraud detectors and extending the features of the toolbox. Some of the planned features are listed in TODO list.

If you use the toolbox in your project, please cite one of the two papers below and the algorithms you used :

CIKM'20 (PDF)

@inproceedings{dou2020enhancing,
  title={Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged Fraudsters},
  author={Dou, Yingtong and Liu, Zhiwei and Sun, Li and Deng, Yutong and Peng, Hao and Yu, Philip S},
  booktitle={Proceedings of the 29th ACM International Conference on Information and Knowledge Management (CIKM'20)},
  year={2020}
}

SIGIR'20 (PDF)

@inproceedings{liu2020alleviating,
  title={Alleviating the Inconsistency Problem of Applying Graph Neural Network to Fraud Detection},
  author={Liu, Zhiwei and Dou, Yingtong and Yu, Philip S. and Deng, Yutong and Peng, Hao},
  booktitle={Proceedings of the 43nd International ACM SIGIR Conference on Research and Development in Information Retrieval},
  year={2020}
}

Useful Resources

Table of Contents

Installation

git clone https://github.com/safe-graph/DGFraud.git
cd DGFraud
python setup.py install

Requirements

* python 3.6, 3.7
* tensorflow>=1.14.0,<2.0
* numpy>=1.16.4
* scipy>=1.2.0
* networkx<=1.11

Datasets

DBLP

We uses the pre-processed DBLP dataset from Jhy1993/HAN You can run the FdGars, Player2Vec, GeniePath and GEM based on the DBLP dataset. Unzip the archive before using the dataset:

cd dataset
unzip DBLP4057_GAT_with_idx_tra200_val_800.zip

Example dataset

We implement example graphs for SemiGNN, GAS and GEM in data_loader.py. Because those models require unique graph structures or node types, which cannot be found in opensource datasets.

Yelp dataset

For GraphConsis, we preprocessed Yelp Spam Review Dataset with reviews as nodes and three relations as edges.

The dataset with .mat format is located at /dataset/YelpChi.zip. The .mat file includes:

  • net_rur, net_rtr, net_rsr: three sparse matrices representing three homo-graphs defined in GraphConsis paper;
  • features: a sparse matrix of 32-dimension handcrafted features;
  • label: a numpy array with the ground truth of nodes. 1 represents spam and 0 represents benign.

The YelpChi data preprocessing details can be found in our CIKM'20 paper. To get the complete metadata of the Yelp dataset, please email to ytongdou@gmail.com for inquiry.

User Guide

Running the example code

You can find the implemented models in algorithms directory. For example, you can run Player2Vec using:

python Player2Vec_main.py

You can specify parameters for models when running the code.

Running on your datasets

Have a look at the load_data_dblp() function in utils/utils.py for an example.

In order to use your own data, you have to provide:

  • adjacency matrices or adjlists (for GAS);
  • a feature matrix
  • a label matrix then split feature matrix and label matrix into testing data and training data.

You can specify a dataset as follows:

python xx_main.py --dataset your_dataset

or by editing xx_main.py

The structure of code

The repository is organized as follows:

  • algorithms/ contains the implemented models and the corresponding example code;
  • base_models/ contains the basic models (GCN);
  • dataset/ contains the necessary dataset files;
  • utils/ contains:
    • loading and splitting the data (data_loader.py);
    • contains various utilities (utils.py).

Implemented Models

Model Paper Venue Reference
SemiGNN A Semi-supervised Graph Attentive Network for Financial Fraud Detection ICDM 2019 BibTex
Player2Vec Key Player Identification in Underground Forums over Attributed Heterogeneous Information Network Embedding Framework CIKM 2019 BibTex
GAS Spam Review Detection with Graph Convolutional Networks CIKM 2019 BibTex
FdGars FdGars: Fraudster Detection via Graph Convolutional Networks in Online App Review System WWW 2019 BibTex
GeniePath GeniePath: Graph Neural Networks with Adaptive Receptive Paths AAAI 2019 BibTex
GEM Heterogeneous Graph Neural Networks for Malicious Account Detection CIKM 2018 BibTex
GraphSAGE Inductive Representation Learning on Large Graphs NIPS 2017 BibTex
GraphConsis Alleviating the Inconsistency Problem of Applying Graph Neural Network to Fraud Detection SIGIR 2020 BibTex
HACUD Cash-Out User Detection Based on Attributed Heterogeneous Information Network with a Hierarchical Attention Mechanism AAAI 2019 BibTex

Model Comparison

Model Application Graph Type Base Model
SemiGNN Financial Fraud Heterogeneous GAT, LINE, DeepWalk
Player2Vec Cyber Criminal Heterogeneous GAT, GCN
GAS Opinion Fraud Heterogeneous GCN, GAT
FdGars Opinion Fraud Homogeneous GCN
GeniePath Financial Fraud Homogeneous GAT
GEM Financial Fraud Heterogeneous GCN
GraphSAGE Opinion Fraud Homogeneous GraphSAGE
GraphConsis Opinion Fraud Heterogeneous GraphSAGE
HACUD Financial Fraud Heterogeneous GAT

TODO List

  • Implementing mini-batch training
  • The log loss for GEM model
  • Time-based sampling for GEM
  • Add sampling methods
  • Benchmarking SOTA models
  • Scalable implementation
  • Pytorch implementation

How to Contribute

You are welcomed to contribute to this open-source toolbox. The detailed instructions will be released soon. Currently, you can create issues or email to bdscsafegraph@gmail.com for inquiry.