Phishing Website Detection Method Based on Multimodal Large Language Model

In recent years, phishing attack methods have demonstrated characteristics of strong anti-detection abilities, resulting in significant economic losses. Traditional phishing website detection methods suffer from limited features and insufficient information, no longer meeting the practical requirements for effective security protection. To address complex and variable network phishing attacks, we propose a phishing website detection method based on multimodal large language models (LLMs). The different modalities of website features are transformed into text features by LLMs. The problem of detecting phishing websites is transformed into a text classification problem. The method we propose improves phishing website detection performance through the multimodal fusion of website information. Experimental evaluation on a public dataset demonstrates the method’s effectiveness, achieving an F1 score of 96.1%.