Researchers have developed PREF-Gate, a novel framework for graph fraud detection that addresses the challenge of fusing label-free graph context with label-derived neighborhood evidence. This auditable decision framework uses two experts and a validation gate to manage conflicting validity conditions between these information sources. Experiments on Amazon, YelpChi, and TFinance datasets demonstrated PREF-Gate's effectiveness, achieving high AUPRC values and showing that label-derived relational evidence is only useful when supported by held-out validation. AI
IMPACT Introduces new methodologies for graph fraud detection, potentially improving security in digital ecosystems by better handling incomplete data and class imbalance.
RANK_REASON The cluster contains two academic papers submitted to arXiv detailing new methods for graph fraud detection.
- alphaXiv
- .amazon
- arXiv
- CatalyzeX
- DagsHub
- GFD-GC
- Gotit.pub
- graph fraud detection
- graph neural networks
- Hugging Face
- PREF-Gate
- ScienceCast
- TFinance
- YelpChi
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
