qw1zzard / gaga Goto Github PK

The PyTorch implementation of GAGA.

Abstract: Node classification is a substantial problem in graph-based fraud detection. Many existing works adopt Graph Neural Networks (GNNs) to enhance fraud detectors. While promising, currently most GNN-based fraud detectors fail to generalize to the low homophily setting. Besides, label utilization has been proved to be significant factor for node classification problem. But we find they are less effective in fraud detection tasks due to the low homophily in graphs. In this work, we propose GAGA, a novel Group AGgregation enhanced TrAnsformer, to tackle the above challenges. Specifically, the group aggregation provides a portable method to cope with the low homophily issue. Such an aggregation explicitly integrates the label information to generate distinguishable neighborhood information. Along with group aggregation, an attempt towards end-to-end trainable group encoding is proposed which augments the original feature space with the class labels. Meanwhile, we devise two additional learnable encodings to recognize the structural and relational context. Then, we combine the group aggregation and the learnable encodings into a Transformer encoder to capture the semantic information. Experimental results clearly show that GAGA outperforms other competitive graph-based fraud detectors by up to 24.39% on two trending public datasets and a real-world industrial dataset from Baidu. Even more, the group aggregation is demonstrated to outperform other label utilization methods (e.g., C&S, BoT/UniMP) in the low homophily setting.

1. Setup.

   • Two public dataset YelpChi and Amazon will be downloaded automatically at first run. Alternatively, you can download both datasets from Github.

   • Requirements

     conda (an open source package management tool) is recommended.

     conda create -n gaga python=3.8
     conda activate gaga
     
     conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
     conda install -c dglteam/label/cu121 dgl
     
     conda install -c conda-forge ruff
     
     pip install numpy pandas scikit-learn tensorboard matplotlib wandb
     
     conda clean --all

2. Prepare sequence data.

   cd preprocessing
   
   # dataset spliting
   python dataset_split.py --dataset yelp  --save_dir seq_data --train_size 0.4 --val_size 0.1
   python dataset_split.py --dataset amazon  --save_dir seq_data --train_size 0.4 --val_size 0.1
   
   # preprocess feature sequence
   python graph2seq_mp.py --dataset yelp --fanouts -1 -1  --save_dir seq_data --train_size 0.4 --val_size 0.1 --n_workers 8 --add_self_loop --norm_feat
   python graph2seq_mp.py --dataset amazon --fanouts -1 -1  --save_dir seq_data --train_size 0.4 --val_size 0.1 --n_workers 8 --add_self_loop --norm_feat

3. Run main_transformer.py

   python main_transformer.py --config configs/yelpchi_paper.json --gpu 0  --log_dir logs --early_stop 100
   python main_transformer.py --config configs/amazon_paper.json --gpu 0  --log_dir logs --early_stop 100

If you use GAGA in a scientific publication, we would appreciate citations to the following paper:

@inproceedings{wang2023label,
  title={Label Information Enhanced Fraud Detection against Low Homophily in Graphs},
  author={Wang, Yuchen and Zhang, Jinghui and Huang, Zhengjie and Li, Weibin and Feng, Shikun and Ma, Ziheng and Sun, Yu and Yu, Dianhai and Dong, Fang and Jin, Jiahui and Wang, Beilun and Luo, Junzhou},
  booktitle={Proceedings of the ACM Web Conference 2023},
  year={2023}
}

Email: [email protected]

GAGA is inspired by the recent success of graph-based fraud detectors (i.e. CARE-GNN, PC-GNN, RioGNN, etc.) and label utilization in node classification tasks (i.e. BoT, UniMP, etc.). We also thank the authors for sharing their codes.