Research on Malicious URL Detection Technology Based on BERT Model

In network security, as malicious URLs increase and change, their detection has gradually become more important. The existing malicious URL detection methods lack the description of location and context semantics. This paper proposes a malicious URL based on the BERT model. The URL detection method first uses the preprocessing method to solve the problem of a large number of random characters forming words in the URL, uses special symbols as a separator to segment the URL, and then trains the BERT model to extract the short string characteristics of the URL and classify it. The experimental results show that the method’s accuracy is 98.30%, the recall rate is 95.21%, and the F1 value is 94.33%.

I. Introduction

With the development of the Internet, the issue of network security has become more and more critical. According to the 47th “Statistical Report on the Development of the Internet in China” issued by the China Internet Network Information Center (CNNIC), as of December 2020, the scale of Chinese Internet users has reached Up to 989 million this number is still increasing. Many users and websites are vulnerable to security threats because of such a large number of users. Various malicious websites, malware, and virus Trojan horses pose considerable threats to users’ privacy and property security. These attacks have evolved to become more and more complex, automated, and challenging to detect.