Scholarlink

Traditional phishing detection often overlooks psychological manipulation. This study investigates using Large Language Model (LLM) In-Context Learning (ICL) for fine-grained classification of phishing emails based on a taxonomy of 40 manipulation techniques. Using few-shot examples with GPT-4o-mini on real-world French phishing emails, we evaluated performance against a human-annotated test set. The approach effectively identifies prevalent techniques (e.g., Baiting, Curiosity Appeal, Request For Minor Favor) with a promising accuracy of 0.76 significantly outperforming traditional machine learning methods including KNN (0.65), Random Forest (0.65), SVM (0.65), Logistic Regression (0.63), and fine-tuned BERT (0.62).

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

In-Context Learning for the Classification of Manipulation Techniques in Phishing Emails

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Antony Dalmiere,
  • Guillaume Auriol,
  • Vincent Nicomette,
  • Pascal Marchand

摘要

Traditional phishing detection often overlooks psychological manipulation. This study investigates using Large Language Model (LLM) In-Context Learning (ICL) for fine-grained classification of phishing emails based on a taxonomy of 40 manipulation techniques. Using few-shot examples with GPT-4o-mini on real-world French phishing emails, we evaluated performance against a human-annotated test set. The approach effectively identifies prevalent techniques (e.g., Baiting, Curiosity Appeal, Request For Minor Favor) with a promising accuracy of 0.76 significantly outperforming traditional machine learning methods including KNN (0.65), Random Forest (0.65), SVM (0.65), Logistic Regression (0.63), and fine-tuned BERT (0.62).