DeepFD-pyTorch

This is a PyTorch implementation of DeepFD (Deep Structure Learning for Fraud Detection), which is used as a baseline method in my paper Error-Bounded Graph Anomaly Loss for GNNs (CIKM20). Other than the unsupervised DBSCAN classifier used in the original paper, I also added a supervised 3-layer MLP as a classifier option. The whole embedding part is still always unsupervised.

Authors of this code package:

Tong Zhao (tzhao2@nd.edu), Kaifeng Yu (ykf16@mails.tsinghua.edu.cn), Chuchen Deng (cdeng@nd.edu).

Environment settings

• python==3.6.8
• pytorch==1.0.1.post2

Basic Usage

Before running the model, first you need to create two folders: results/ and data/.

Data Inputs

Required input data files are graph_u2p and labels, the paths need to be modified in configs/file_paths.json.

graph_u2p is the pickled adjacency matrix in scipy.sparse.csr_matrix format, where each none-zero entry stands for a edge.

labels is the pickled binary labels in numpy.ndarray format, where 1 stands for fraudulent user and 0 stands for benign user. For dataset with limited labels, the unlabeled user should be labeled as -1 in the labels vector.

Example Usage

To run the unsupervised model on Cuda with the default GPU card:

python -m src.main --cuda 9 --dataSet [YourDataSet] --cls_method [dbscan or mlp]

Main Parameters:

--dataSet     The input graph dataset. (default: weibo_s)
--name        The name of this run. (default: debug)
--cls_method  The classification method to be used. Choose between dbscan and mlp. (default: dbscan)
--epochs      Number of epochs. (default: 10)
--b_sz        Batch size. (default: 100)
--seed        Random seed. (default: 1234)
--hidden_size The size of hidden layer in encoder and decoder. (default: 128)
--emb_size    The size of the embeddings for each user. (default: 2)
--cuda        Which GPU card to use. -1 for CPU, 9 for default GPU, 0~3 for specific GPU. (default: -1)

Cite

If you find this repository useful in your research, please cite our paper:

@inproceedings{zhao2020error,
  title={Error-Bounded Graph Anomaly Loss for GNNs},
  author={Zhao, Tong and Deng, Chuchen and Yu, Kaifeng and Jiang, Tianwen and Wang, Daheng and Jiang, Meng},
  booktitle={Proceedings of the 29th ACM International Conference on Information and Knowledge Management},
  pages={},
  year={2020}
}