A Novel Semantic-Aware Approach for Detecting Malicious Web Traffic

摘要

With regard to web compromise, malicious web traffic refers to requests from users visiting websites for malicious targets, such as web vulnerabilities, web shells and uploaded malicious advertising web pages. To directly and comprehensively understand malicious web visits is meaningful to prevent web compromise. However, it is challenging to identify different malicious web traffic with a generic model. In this paper, a novel semantic-aware approach is proposed to detect malicious web traffic by profiling web visits individually. And a semantic representation of malicious activities is introduced to make detection results more understandable. The evaluation shows that our algorithm is effective in detecting malice with an average precision and recall of 90.8% and 92.9% respectively. Furthermore, we employ our approach on more than 136 million web traffic logs collected from a web hosting service provider, where 3,995 unique malicious IPs are detected involving hundreds of websites. The derived results reveal that our method is conductive to figure out adversaries' intentions.