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

Graph anomaly detection (GAD) is a vital task in graph-based machine learning and has been widely applied many real-world applications. The primary goal of GAD to capture anomalous nodes from graph datasets, which evidently deviate the majority nodes. Recent methods have paid attention various scales contrastive strategies for GAD, i.e., node-subgraph node-node contrasts. However, they neglect subgraph-subgraph comparison information normal abnormal subgraph pairs behave differently terms embeddings structures resulting sub-optimal performance. In this paper, we fulfill above idea proposed multi-view multi-scale framework with contrast first practice. To be specific, regard original input as view generate second by augmentation edge modifications. With guidance maximizing similarity pairs, contributes more robust despite structure variation. Moreover, introduced cooperates well widely-adopted counterparts mutual performance promotions. Besides, also conduct sufficient experiments investigate impact different approaches on comprehensive experimental results demonstrate superiority our method compared state-of-the-art effectiveness pair strategy task. source code released at https://github.com/FelixDJC/GRADATE.