A Malicious Android Malware Detection System based on Implicit Relationship Mining

摘要

Nowadays, Android system is the most popular mobile smart operating system on the market. However, the number of Android malware has increased sharply and Android malware spreads quickly and has lots of variants. The existing traditional Android malware detection systems are mostly based on the explicit relationship between Android applications, and their detection accuracy and detection efficiency need to be improved. In this paper, we establish an heterogeneous information network to represent the structural and semantic relations between Android application entities. Then we combine relation matrices and meta-paths to get entity features, and we propose two aggregation operations to learn Android application features. Finally, we use the Multi-Layer Perception neural network to detect Android malware. Through testing on the real data sets, the results show that our system takes into account both accuracy and efficiency.