Скачать или смотреть KDD 2023 - Removing Camouflage, Revealing Collusion: Leveraging Gang-crime Fraudster Detection

Lewen Wang, Microsoft

Gang crime is common in fraud scenarios. Researchers have explored both temporal and spatial collusive patterns to detect fraudsters within gangs. However, fraud gangs often engage in community-level temporal and spatial camouflage, concealing collusive patterns to appear as benign communities. This camouflage poses challenges for detection methods. Additionally, existing approaches struggle with sample imbalance caused by rare fraudsters hidden among numerous users. To address these issues, we propose a GAN-based framework, named ACD. ACD comprises four modules, in charge of community division, camouflage identification, fraudster detection, and camouflage generation. The first three modules work together as a discriminator to enhance fraudster detection by identifying and eliminating camouflage. The camouflage generation module serves as the generator, competing against the discriminator to generate fraudster samples. This approach mitigates the challenge of sample imbalance and improves model robustness.