Graph Anomaly Detection via Multi-Scale Contrastive Learning Networks with Augmented View

Duan, Jingcan; Wang, Siwei; Zhang, Pei; Zhu, En; Hu, Jingtao; Hu, Jin; Liu, Yue; Dong, Zhibin

doi:10.48550/arxiv.2212.00535

Cited by 1 publication

(11 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Multi-scale contrast in GCAD can be abstracted as node-node, subgraph-subgraph, and node-subgraph contrasts, which focus on different interaction patterns. By summarising the previous GCAD methods [4,7,14,33], we noticed that graph G can generate three type views of node v i , subgraph features s i , node features Z i , and masked node features (Z s i ) i as shown in Fig. 1.…”

Section: Multi-scale Contrastmentioning

confidence: 90%

“…It's natural to consider that subgraph-subgraph contrast shall be a promising idea for modeling more complex interaction patterns. GRADATE [4] constructed multi-scale contrasts, including nodenode, subgraph-node, subgraph-subgraph. To capture more comprehensive graph level representations, Luo et.al [17] designed a new graph level evaluation metrics and built an end-to-end anomaly detection framework based on contrastive learning.…”

Section: Contrastive-basedmentioning

confidence: 99%

“…The overview framework of our MAG, which unified the CoLA [14], ANEMONE [7] and GRADATE [4] via contrast combinations from top to bottom in the combination pool. The MAG framework consists of two modules: graph augmentation and multi-GNN modules.…”

Section: Node-node Contrastmentioning

confidence: 99%

“…To address this issue, contrastive learning, benefited from its self-supervised property, has been combined with GNN models for anomaly detection. Existing graph contrastive anomaly detection (GCAD) methods, such as ANEMONE [7], SL-GAD [33], and GRADATE [4], have utilized graph augmentation or multi-scale contrast modules to upgrade their models. However, these incremental works enhance the expression of the model by adding different multi-scale contrasts or graph augmentation strategies intuitively without any empirical design guidance.…”

Section: Introductionmentioning

confidence: 99%

“…We traversed thoroughly the single contrast instances of the MAG framework and observed that the normal node-subgraph contrast had better detection performance than the node-node, sugraph-subgraph, and masked node-subgraph contrasts. Unlike the GRADATE [4] model used a variety of multi-scale contrast combinations, our lightweight L-MAG surpasses the state-of-the-art on Cora and Pubmed with the low computational cost. The variant M-MAG model equipped with multi-GNN modules further improve the detection performance.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Revisiting Graph Contrastive Learning for Anomaly Detection

Liu,

Cao,

Tao

et al. 2023

Frontiers in Artificial Intelligence and Applications

View full text Add to dashboard Cite

Combining Graph neural networks (GNNs) with contrastive learning for anomaly detection has drawn rising attention recently. Existing graph contrastive anomaly detection (GCAD) methods have primarily focused on improving detection capability through graph augmentation and multi-scale contrast modules. However, the underlying mechanisms of how these modules work have not been fully explored. We dive into the multi-scale and graph augmentation mechanism and observed that multi-scale contrast modules do not enhance the expression, while the multi-GNN modules are the hidden contributors. Previous studies have tended to attribute the benefits brought by multi-GNN to the multi-scale modules. In the paper, we delve into the misconception and propose Multi-GNN and Augmented Graph contrastive framework MAG, which unified the existing GCAD methods in the contrastive self-supervised perspective. We extracted two variants from the MAG framework, L-MAG and M-MAG. The L-MAG is the lightweight instance of the MAG, which outperform the state-of-the-art on Cora and Pubmed with the low computational cost. The variant M-MAG equipped with multi-GNN modules further improve the detection performance. Our study sheds light on the drawback of the existing GCAD methods and demonstrates the potential of multi-GNN and graph augmentation modules. Our code is available at https://anonymous.4open.science/r/MAG-Framework-74D0.

show abstract

Section: Multi-scale Contrastmentioning

confidence: 90%

Section: Contrastive-basedmentioning

confidence: 99%