CATeye identifies voucher abuse by learning which data features remain stable as fraud patterns evolve.

Online stores struggle when fraudsters abuse promotional discount vouchers to make unfair profits. These fraud tricks change quickly over time and across different regions. When user behaviors shift, the connections built between similar users shift as well. This confuses standard graph-based detection systems. To solve this, researchers designed a framework called CATeye. It works by separating stable clues from misleading, temporary changes. First, a tool called the Attribute Invariance Selector filters out user traits that shift across settings. Next, conditioned on the remaining stable traits, an Edge Invariance Selector keeps reliable connections between users while setting aside shifting connections. Finally, CATeye builds multiple data views to focus on stable signals and ignore setting-specific noise. Across tests, it consistently outperformed nine strong detection methods, boosting detection accuracy by up to 8.61%.

Why it matters

E-commerce platforms lose money to voucher fraud, and current models fail when fraud tactics change. This framework helps detect abuse without needing constant retraining, saving resources and improving security.

Questions answered