APFormer: Anti-Phishing Transformer for Website-Phishing Detection Via Joint Feature Learning

Phishing is a cybercrime activity where the perpe-trator tricks users into believing that a fraudulent website is authentic, revealing sensitive information. Recently, the frequency of phishing attacks has increased, targeting both individuals and corporate sectors, thereby rendering traditional methods of detecting phishing websites less effective. These methods often produce high false positive rates due to the complexities involved in creating a strong correlation between brand identity and website features. Furthermore, sophisticated attacks, such as DOM element manipulation tricks, evade the existing methods and convince the user with visual and behavioural similarity. We propose a transformer-based approach for website phishing detection using joint feature learning. The approach jointly learns textual, visual, and heuristic multi-modal features to establish a stronger connection between brand identity and the webpage. The proposed model comprises a feature extraction module and a transformer network. The feature extraction module extracts textual, visual, and heuristic features from the DOM tree, while the transformer module classifies the input page as legitimate or phishing by learning the feature mapping to brand identity. Our method outperforms current state-of-the-art methods on benchmark datasets by achieving a 3.5% increase in recall and a 5.9% reduction in false positive rate.