Beyond Homophily: Robust Graph Anomaly Detection via Neural Sparsification
Authors: Zheng Gong, Guifeng Wang, Ying Sun, Qi Liu, Yuting Ning, Hui Xiong, Jingyu Peng
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on real-world datasets of GAD demonstrate that the proposed framework achieves significantly better detection quality compared with the state-of-the-art methods, even when the graph is heavily attacked. |
| Researcher Affiliation | Collaboration | Zheng Gong1,2 , Guifeng Wang3, , Ying Sun4 , Qi Liu1,2 , Yuting Ning1,2 , Hui Xiong4 and Jingyu Peng1,2 1School of Computer Science and Technology, University of Science and Technology of China 2State Key Laboratory of Cognitive Intelligence 3Huawei Technologies Co Ltd 4Hong Kong University of Science and Technology (Guangzhou) |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. The methods are described using mathematical equations and textual explanations. |
| Open Source Code | Yes | Code will be available at https://github.com/Kelly Gong/Sparse GAD.git. |
| Open Datasets | Yes | We conduct extensive experiments to evaluate Sparse GAD on three real-world anomaly detection datasets: Amazon [Mc Auley and Leskovec, 2013], Yelp Chi [Rayana and Akoglu, 2015], Reddit [Kumar et al., 2019]. |
| Dataset Splits | Yes | We select 40% nodes for training models in supervised scenarios and 1% in semi-supervised scenarios. The remaining nodes are split by 1:2 for validation and testing. |
| Hardware Specification | No | The paper mentions accelerating kNN "on GPU devices" but does not specify any particular GPU models, CPU types, or other detailed hardware specifications used for the experiments. |
| Software Dependencies | No | The paper mentions using "Adam optimizer" and loading the Reddit dataset from the "PyGod package [Liu et al., 2022a]", but it does not specify version numbers for general software dependencies like Python, PyTorch/TensorFlow, or other libraries. |
| Experiment Setup | Yes | For each experiment, we train all models for 2000 epochs by Adam optimizer. We optimize all models on each dataset by selecting learning rate from {0.01, 0.005, 0.001}, hidden states from {16, 32, 64} and dropout rate from {0, 0.1, 0.2} via grid search, and save the model according to the best AUC in validation. As for Sparse GAD, α of Equation (2) and λ are both set to 0.1, and δ is set to 0.05 for all datasets. For each dataset, we search k of k NN from {5, 10, 20} and τ from {0.5, 1.0, 2.0}. |