ABCF: An Adaptive Balanced Multimodal Website Classification Framework

Websites are a crucial medium for conveying multimedia information. However, malicious websites can lurk among them, posing a threat to the security of users’ sensitive data and personal privacy. Hence, it is significant to identify and categorize such harmful sites. Various solutions have been proposed by researchers for this task, including extracting features from URLs and analyzing web page content using classification algorithms. To tackle these challenges, we suggest the Adaptive Balanced multimodal website Classification Framework (ABCF), which is designed for web page classification tasks. The framework includes a FAModule and FEAModule for better multimodal data fusion on websites. We also employ incremental learning to continuously improve our model and adapt to the changing network environment. Experiments on three website datasets confirm the effectiveness of our modules in capturing website features and demonstrate the superiority of our model compared to past malicious website classification systems. Code will be released at github.com/lzyy2435/ABCF/.