Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Revisiting Graph Contrastive Learning on Anomaly Detection: A Structural Imbalance Perspective
Authors: Yiming Xu, Zhen Peng, Bin Shi, Xu Hua, Bo Dong, Song Wang, Chen Chen
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The performance evaluation of the whole, head, and tail nodes on multiple datasets validates the comprehensive superiority of the proposed AD-GCL in detecting both head anomalies and tail anomalies. Experiments: Comprehensive performance evaluation of the whole, head, and tail nodes on six public datasets demonstrates the superiority of our proposed AD-GCL over the baselines. Main Results and Analysis We evaluate the anomaly detection performance of AD-GCL by conducting a comparison with 10 baseline methods. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, Xi an Jiaotong University 2Shaanxi Provincial Key Laboratory of Big Data Knowledge Engineering, Xi an Jiaotong University 3School of Distance Education, Xi an Jiaotong University 4University of Virginia, Charlottesville, Virginia, USA EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology using prose and mathematical equations but does not contain any explicit sections or blocks labeled "Pseudocode" or "Algorithm". |
| Open Source Code | Yes | Code https://github.com/yimingxu24/AD-GCL |
| Open Datasets | Yes | To conduct a comprehensive comparison, we evaluate AD-GCL on six widely used benchmark datasets for anomaly detection. Specifically, we choose two categories of datasets: 1) citation networks (Liu et al. 2021b; Jin et al. 2021) including Cora, Citeseer, and Pubmed, 2) bitcoin trading networks (Kumar et al. 2016, 2018) including Bitcoinotc, BITotc, and BITalpha. |
| Dataset Splits | No | The paper defines how head and tail nodes are categorized based on a degree threshold K, and mentions evaluating performance on these categories. However, it does not provide specific information regarding training, validation, and test dataset splits (e.g., percentages or sample counts) for model training and evaluation. |
| Hardware Specification | No | The paper states, "Please refer to the Appendix for more detailed parameter settings and experimental environment introduction," but no specific hardware details (like GPU/CPU models or memory) are provided in the main text. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions) needed to replicate the experiments. |
| Experiment Setup | Yes | Parameter Study: Effect of sampling rounds R We investigate the effect of varying the value of R in Eq. (10), as shown in Figure 4a. Effect of hidden layer dimension d We visualize the effect of the hidden layer dimension d in Figure 4b. Effect of trade-off parameter α In Figure 4c, we present the evaluation results for different values of α in Eq. (9). Effect of sliding window w We study the sliding window w in the neighbor completion strategy, which incorporates anomaly scores from the previous w training rounds to compute the anomaly similarity distribution. Figure 4d shows the best performance is achieved when w = 5... |