Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Rethinking Feature Alignment in Generalist Graph Anomaly Detection: A Relational Fingerprint-based Approach

About

Generalist graph anomaly detection (GAD) aims to detect anomalies on unseen graphs without graph-specific retraining. Nevertheless, existing approaches primarily focus on aligning heterogeneous features across different data domains via PCA-based projection, which harmonizes feature dimensions ignores feature semantics. As a result, GAD models fail to learn transferable semantic knowledge, and even exhibit negative transfer on unseen graphs. To address this issue, we propose a Relational Fingerprint-based generalist GAD approach (ReFi-GAD for short), aligning heterogeneous raw features with a universal and semantics-aware Relational Fingerprint (ReFi) that encodes anomaly-indicative cues from both contextual and structural perspectives. Building on ReFi, we design a fingerprint-grounded generalist GAD model, which combines a transformer-based encoder to capture domain-invariant knowledge with an SNR-guided refinement module for domain-specific adaptation. Extensive experiments on 14 datasets demonstrate that ReFi-GAD significantly outperforms state-of-the-art methods.

Yujing Liu, Yixin Liu, Yu Zheng, Alan Wee-Chung Liew, Xiaofeng Cao, Shirui Pan• 2026

Related benchmarks

TaskDatasetResultRank
Graph Anomaly DetectionAMAZON
AUROC71.21
109
Graph Anomaly DetectionREDDIT
AUROC62.11
106
Graph Anomaly DetectionBlogCatalog
AUROC0.7679
101
Graph Anomaly DetectionWeibo
AUROC92.66
99
Graph Anomaly DetectionFacebook
AUROC0.9467
75
Graph Anomaly DetectionPubmed
AUC97.66
65
Graph Anomaly Detectionquestions
AUPRC4.55
59
Graph Anomaly DetectionACM
AUPRC0.6215
54
Graph Anomaly DetectionCora--
50
Graph Anomaly DetectionYelpChi
AUROC83.4
49
Showing 10 of 17 rows

Other info

Follow for update