Machine Learning-Based Detection of Gambling-Like Mobile Applications and Their Resource Coordination

Mobile gambling applications constitute a significant component of the online black and grey industries. They lure users into illegal activities through covert distribution channels, causing substantial financial losses and social harm. To evade detection, these applications often employ techniques like code obfuscation, frequent domain changes, and rotating developer certificates, rendering traditional detection methods based on single-application features less effective. This paper proposes a collaborative detection framework based on machine learning, innovatively integrating multimodal contrastive learning and Heterogeneous Graph Neural Networks (HGNN) to achieve precise identification of gambling applications and uncover their ecosystems. Firstly, we collect mobile application samples on a large scale and utilize deep learning techniques (e.g., Transformer) to learn deep feature representations from multimodal data including APKs, domains, and certificates. Secondly, we design an association relationship modeling method based on contrastive learning to automatically learn the optimal similarity metric between gambling applications and their resources. Furthermore, we construct a heterogeneous information network comprising four types of nodes: applications, SDKs, domains, and certificates. We apply a Heterogeneous Graph Attention Network (HAN) for end-to-end representation learning and community discovery, accurately identifying gambling application families and their shared resource networks.