Pré-Publication, Document De Travail Année : 2026

In-context learning for the classification of manipulation techniques in phishing emails

Résumé

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).

Fichier principal
Vignette du fichier
main3.pdf (1.05 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05127120 , version 1 (25-06-2025)

Licence

Identifiants

Citer

Antony Dalmiere, Guillaume Auriol, Vincent Nicomette, Pascal Marchand. In-context learning for the classification of manipulation techniques in phishing emails. 2025. ⟨hal-05127120⟩
1663 Consultations
211 Téléchargements

Altmetric

Partager

  • More