Phishing email uniform resource locator detection based on large language model

With the advent of digital transformation and the proliferation of online services, a myriad of phishing attacks against individual users and organizations emerge incessantly. Attackers employ increasingly sophisticated covert techniques to create phishing links, including the use of uniform resource locator (URL) shortening services, dynamic URL generation, and the utilization of legitimate websites as vectors. Many phishing attacks also incorporate social engineering tactics, raising the risk of victim deception by creating a sense of urgency or enticing users to click on links, such as those posing as notifications from banks, social media, or critical institutions. Phishing URLs are primarily propagated via email, social media, instant messaging, and text messages, with malicious phishing email URLs currently exhibiting widespread dissemination. This paper employs the Qwen 2-72B model to detect malicious phishing email URLs. Equipped with natural language processing capabilities, the model can thoroughly comprehend the context and content of Chinese emails, enabling it to more precisely identify potential malicious links. Furthermore, the model, trained on extensive data, can identify a variety of URL characteristics and phishing tactics, thereby enhancing detection accuracy. Additionally, the model provides real-time analysis of URLs within emails, rapidly detecting potential phishing links and thereby significantly enhancing user security. Utilizing deep learning algorithms, the model boasts rich semantic understanding capabilities, enabling it to reduce false positives and elevate the detection rate of malicious link identifications. Experimental results demonstrate that the model exhibits high detection accuracy in identifying malicious URLs within Chinese emails, thereby strengthening network security defenses and enhancing user protection.