Novel UGA Homologous URL Recognition in Real-World Financial Cybercrimes: Self-supervised Deep Learning of URL Semantics

Financial cybercrime poses a growing concern for financial institutions, requiring measures to identify and block illegal payments. In combating cybercriminals on Alipay, we’ve discovered a novel and previously unreported type of malicious URL, which algorithmically generated with random strings. Besides, it is crucial to note that even homologous URLs have significant differences in their text content, but only share certain similarities in structural patterns, thereby enabling them to evade detection successfully. Recognizing these novel homologous URLs presents challenges due to their inherent lack of awareness, absence of labeled data, and limited textual similarity. To address this, we propose a self-supervised learning approach utilizing the Deep Quadruplet Siamese Neural Network (DQSN) to learn the representation of URL structure and abstract semantics. Our approach yields promising results on Alipay, demonstrating its remarkable ability to identify even previously unseen URL patterns.