Contextual Analysis by a BERT-Based Approach for Intelligent Detection of Dictionary-Generated Malicious Domain

Abstract

The increasing use of domain generation algorithms (DGAs) in malicious attacks represents a major challenge for modern cybersecurity. Dictionary-based DGAs, which generate semantic domain names, render traditional detection methods ineffective. This paper proposes a specific model for the accurate detection of these domains. Our approach introduces targeted pre-training on corpora of malicious domain names, accompanied by novel selective attention strategies. Experimental results demonstrate a significant improvement in accuracy and reduction in inference time, positioning our model as a lightweight yet powerful solution for network security.