A Novel Malware Detection Framework Based on Weighted Heterograph
Meihua Fan, Shudong Li, Weihong Han, Xiaobo Wu, Zhaoquan Gu, Zhihong Tian · 2020
Malware has always been one of the major threats to cyberspace security. Also, it has long been concerned by researchers of antivirus technology. However, malware gradually tends to be diversified and updated quickly, it is a big challenge to security personnel both in terms of number and attack means. In this paper, we propose a malware detection framework based on fine-grained malware behavior graph, which applies graph neural network to malware detection tasks. We design a fine-grained malicious behavior graph to represent the association of malware and its behavior associated entities. Then, aiming at learning the semantic information carried in the malicious behavior graph, we propose a weighted heterogeneous graph neural network named MalSage. We conducted a model evaluation of the proposed method on a malware dataset from VirusTotal. The results show that the proposed framework is more accurate than other algorithms in malware classification task.