Cross-Domain Graph Anomaly Detection via Anomaly-Aware Contrastive Alignment

Authors: Qizhou Wang, Guansong Pang, Mahsa Salehi, Wray Buntine, Christopher Leckie

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on eight CD-GAD settings demonstrate that our approach ACT achieves substantially improved detection performance over 10 state-of-the-art GAD methods.
Researcher Affiliation Academia 1 Monash University 2 Singapore Management University 3 Vin University 4 The University of Melbourne
Pseudocode No The paper describes its method in text and equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/QZ-WANG/ACT.
Open Datasets Yes Eight CD-GAD settings based on four real-world GAD datasets, including Yelp Hotel (HTL), Yelp Res (RES), Yelp NYC (NYC) and Amazon (AMZ)1, are created... 1Statistics of each dataset are given in Suppl. Material
Dataset Splits No The paper describes the datasets used (Yelp Hotel, Yelp Res, Yelp NYC, Amazon) and states '1Statistics of each dataset are given in Suppl. Material' but does not provide specific train/validation/test split percentages or sample counts in the main text.
Hardware Specification No The paper mentions running experiments on the 'MASSIVE HPC facility', but does not provide specific details such as GPU or CPU models, or other hardware specifications.
Software Dependencies No The paper mentions the use of 'Graph SAGE' and 'ADAM optimiser' but does not specify version numbers for any software dependencies.
Experiment Setup Yes Our model ACT is implemented with a three-layer Graph SAGE ... 256 and 64 hidden dimensions are chosen for ψs and ψt respectively. The source model is trained for 50 epochs using a learning rate of 10 3. The domain alignment is performed for 50 epochs using a learning rate of 10 4. ... The optimisation is done in mini-batches of 128 target (centre) nodes using the ADAM optimiser... We use the sample size of 25 and 10 for the two hidden layers during message passing. In self labelling, α = 2.5 and q = 25 are used by default.