Revisiting Graph-Based Fraud Detection in Sight of Heterophily and Spectrum
Authors: Fan Xu, Nan Wang, Hao Wu, Xuezhi Wen, Xibin Zhao, Hai Wan
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The comprehensive experimental results on four real-world fraud detection datasets denote that SEC-GFD outperforms other competitive graph-based fraud detectors. We release our code at https://github.com/Sunxkissed/SEC-GFD. Experiments In this section, we evaluate the performance of our proposed approach on real-world datasets through a series of experiments, and compare the results with those of state-of-the-art baselines to demonstrate the effectiveness of our approach. In addition, we conduct experiments on the response of heterophily edges to distinct filters, as well as experiments on ablation and hyperparameter analysis. |
| Researcher Affiliation | Academia | Fan Xu3, Nan Wang1*, Hao Wu3, Xuezhi Wen1, Xibin Zhao2*, Hai Wan2 1 School of Software, Beijing Jiaotong University, Beijing, China 2 BNRist, KLISS, School of Software, Tsinghua University, Beijing, China 3 IAT, University of Science and Technology of China, Hefei, China {markxu, wuhao2022}@mail.ustc.edu.cn; {wangnanbjtu, 22126399}@bjtu.edu.cn; {zxb, wanhai}@tsinghua.edu.cn |
| Pseudocode | No | The paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | We release our code at https://github.com/Sunxkissed/SEC-GFD. |
| Open Datasets | Yes | We conduct experiments on four real-world datasets targeted at fraud detection scenarios. Overall, they are Amazon (Mc Auley and Leskovec 2013), Yelp Chi (Rayana and Akoglu 2015), and two recently published transaction datasets, i.e., T-Finance and T-Social (Tang et al. 2022). |
| Dataset Splits | Yes | In the experiments, to ensure fairness, the size of training/validation/testing set of the datasets is set to 0.4/0.2/0.4 for all the compared methods. |
| Hardware Specification | No | The paper does not specify any hardware details such as GPU models, CPU types, or memory used for the experiments. |
| Software Dependencies | No | The paper states: "Our method is implemented with the DGL library in Pytorch", but it does not specify version numbers for DGL or Pytorch. |
| Experiment Setup | Yes | In our proposed method SEC-GFD, for the Amazon, Yelp Chi, and T-Finance datasets, the hidden layer dimension is set to 64, and the high-frequency signal neighbor order C is set to 2. For the T-Social dataset, the hidden layer dimension is set to 16, the order C of high-frequency signal neighbors is set to 5. Furthermore, we run for 100 epochs on the four datasets for all methods. In order to better combine the two loss functions, a hyperparameter α [0, 1] is added to balance their influences. |