An Approach for Business Email Compromise Detection using NLP and Deep Learning

Business email compromise is a tool used by adversaries to attack various organizations. Detecting such cyberattacks using well-known methods is becoming increasingly difficult. This paper proposes an approach based on natural language processing and deep neural networks, such as BiGRU and CNN, to detect business email compromise attacks. Semantic features are extracted from emails using a pre-trained BERT model. At the same time, the BiGRU and CNN models allow local feature extraction from emails. Three datasets of different sizes containing phishing and legitimate emails are considered for the experiments. Comparison results with other well-known methods demonstrate the applicability of the proposed Hybrid BERT+BiGRU+CNN model, outperforming them and showing accuracy of 99.59%, 98.77%, and 98.67% on the Ling-Spam, Enron-Spam, and TREC 2007 datasets, respectively. The proposed approach is a tool for business email compromise detection, providing various organizations with an effective solution against cyberattacks.