TriMNet-P: Tri-Modality Network for Phishing Attack Detection on Social and Digital Communication Platforms

Phishing attacks are increasingly spreading through social media channels using malicious URLs, spoofed websites, and malicious attachments that capture personal information. Conventional detection systems are at a loss when detecting such multimodal attacks because they are based on single-modality analysis. To overcome detection of such attacks, the proposed TriMNet-P (Tri-Modality Network for Phishing Attack Detection) uses deep learning to combine detection of textual, visual, and network behavioural features. A Context-Aware Modality Attention (CAMA) mechanism dynamically weights relevant inputs across email, WhatsApp, and Instagram platforms. Tested on Phishing Email, Open Phish, and CIC-IDS2017 datasets, TriMNet-P has attained 98.6% accuracy, 97.9% precision, 97.5% recall, and runs at only 6ms latency.