TERGAD: LLM-Based Graph Anomaly Detection Framework
Researchers propose TERGAD, a novel data augmentation framework that uses Large Language Models (LLMs) to enhance graph anomaly detection by translating node-level topological properties into natural language narratives. The approach addresses limitations in existing text-rich methods that neglect structural context, enabling detection of anomalies arising from inconsistencies between node content and topological role. The framework is detailed in arXiv paper 2605.19738.
Key facts
- TERGAD stands for Structure-aware Text-enhanced Representations for Graph Anomaly Detection
- It uses LLMs to generate descriptive narratives from node topology
- Addresses limitations of existing text-rich GAD methods
- Detects anomalies from content-topology inconsistencies
- Published on arXiv with ID 2605.19738
- Announce type is cross
- Focuses on nodes, edges, or substructures that deviate from majority
- Novel data augmentation framework for GAD
Entities
Institutions
- arXiv