Fewshing: A Few-Shot Learning Approach to Phishing Email Detection

Phishing email attacks persistently present challenges to the privacy and property of the public. To address the challenge posed by phishing emails, many existing machine learning methods, which utilize email content as features, have been proposed. However, insufficient attention has been given to the evolving nature of phishing email content and its impact on the performance of detection models. Moreover, acquiring enough up-to-date phishing email samples, suitable for learning the evolving content, is also a challenge. This paper analyzes the evolution of phishing email content and explores its impact on the performance of detection models. It proposes Fewshing, a few-shot learning approach to detect phishing emails. The experimental results show that Fewshing achieves an F1 score of 92.4% and an accuracy of 98.6% on the limited and imbalanced training datasets, which demonstrates the effectiveness of Fewshing in detecting phishing emails.