CAMERA: Adapting to Semantic Camouflage in Unsupervised Text-Attributed Graph Fraud Detection

Researchers introduced CAMERA, an unsupervised fraud-detection framework for text-attributed graphs.

It is designed to catch camouflaged fraudsters who imitate benign text to evade standard detectors.

CAMERA uses expert modules and context-aware gating to fuse multiple fraud cues in a one-class setting.

The approach aims to strengthen fraud protection for online and e-commerce platforms.