Phishing Domain Name Detection Based on Hierarchical Fusion of Multimodal Features

Phishing attack could trick a victim into revealing sensitive personal, corporate, or financial information through various technical means. As a typical kind of cybercrimes, phishing attack has become one of the major security risks on the Internet leading to tremendous losses to the whole community. Note that phishing attackers could evade detection by multiple ways, for instance, by simply replacing the text with some image and by using slightly rotated or deformed logo images. To address these above limitations, in this paper we proposed a new phishing detection procedure based on hierarchical fusion features from multiple aspects, including domain name, webpage text, webpage structure and webpage logo image. We validated the above procedure through experiments, which indicates that our solution could detect phishing domain names (or websites) more effectively than those traditional ways.