Abstract:Graph anomaly detection has attracted considerable attention from various domain ranging from network security to finance in recent years. Due to the fact that labeling is very costly, existing methods are predominately developed in an unsupervised manner. However, the detected anomalies may be found out uninteresting instances due to the absence of prior knowledge regarding the anomalies looking for. This issue may be solved by using few labeled anomalies as prior knowledge. In real-world scenarios, we can easily obtain few labeled anomalies. Efficiently leveraging labelled anomalies as prior knowledge is crucial for graph anomaly detection; however, this process remains challenging due to the inherently limited number of anomalies available. To address the problem, we propose a novel approach that leverages metapath to embed actual connectivity patterns between anomalous and normal nodes. To further efficiently exploit context information from metapath-based anomaly subgraph, we present a new framework, Metapath-based Graph Anomaly Detection (MGAD), incorporating GCN layers in both the dual-encoders and decoders to efficiently propagate context information between abnormal and normal nodes. Specifically, MGAD employs GNN-based graph autoencoder as its backbone network. Moreover, dual encoders capture the complex interactions and metapath-based context information between labeled and unlabeled nodes both globally and locally. Through a comprehensive set of experiments conducted on seven real-world networks, this paper demonstrates the superiority of the MGAD method compared to state-of-the-art techniques. The code is available at <a class="link-external link-https" href="https://github.com/missinghwan/MGAD" rel="external noopener nofollow">this https URL</a>.

Graph Anomaly Detection Via Prototype-Aware Label Propagation (student Abstract)

Accurate Anomaly Detection Leveraging Knowledge-enhanced GAT

LabelGen: An Anomaly Label Generative Framework for Enhanced Graph Anomaly Detection

DAGAD: Data Augmentation for Graph Anomaly Detection

GADY: Unsupervised Anomaly Detection on Dynamic Graphs

Graph Anomaly Detection with Diffusion Model-Based Graph Enhancement (student Abstract)

A Universal Adaptive Algorithm for Graph Anomaly Detection

Label-based Graph Augmentation with Metapath for Graph Anomaly Detection

Graph Anomaly Detection Via Multi-View Discriminative Awareness Learning

Multi-representations Space Separation based Graph-level Anomaly-aware Detection

Prototype-oriented Hypergraph Representation Learning for Anomaly Detection in Tabular Data

An Anomaly Detection Method Based on Meta-Path and Heterogeneous Graph Attention Network

Self-supervised End-to-end Graph Level Anomaly Detection

MixedAD: A Scalable Algorithm for Detecting Mixed Anomalies in Attributed Graphs.

Graph-level Anomaly Detection via Deep Metric Learning.

Generative Semi-supervised Graph Anomaly Detection

Improving Generalizability of Graph Anomaly Detection Models Via Data Augmentation

Generative and Contrastive Self-Supervised Learning for Graph Anomaly Detection

Counterfactual Graph Learning for Anomaly Detection on Attributed Networks

Alleviating Structural Distribution Shift in Graph Anomaly Detection

Graph Anomaly Detection at Group Level: A Topology Pattern Enhanced Unsupervised Approach