Phishing Email Detection Using Improved RCNN Model With Multilevel Vectors and Attention Mechanism

Metadane

• Autorzy: Y. Fang, C. Zhang, C. Huang, L. Liu, Y. Yang
• Rok: 2019
• Źródło: IEEE Access, Volume 7, Pages 56329–56340
• DOI/Link: 10.1109/ACCESS.2019.2913705
• Status: to-read
• Pochodzenie: Wyekstrahowane z al-subaiey-web-ai-phishing-2024 ([33] - RCNN 99% accuracy, Table 3 comparison)
• Tagi: to-read reference rcnn attention-mechanism word2vec phishing-email deep-learning multilevel-vectors

Notatki

Publikacja dodana automatycznie z bibliografii.

Kontekst cytowania w Al-Subaiey 2024:

• Referenced jako [33] w paper (Table 1, Table 3 comparisons)
• Method: RCNN (Recurrent Convolutional Neural Network) using:
  • Multilevel vectors
  • Attention mechanisms
  • Word2Vec embeddings (email body + header data)
• Dataset: Unspecified (not disclosed in Al-Subaiey citation)
• Results: Accuracy 99.00%
• Comparison: Matches Al-Subaiey performance (99.1% vs 99.0%) but with complex architecture

Technical Sophistication:

• RCNN combines CNN (spatial features) + RNN (sequential patterns)
• Attention mechanism: focuses on relevant email parts
• Multilevel vectors: different granularities of text representation
• Word2Vec: semantic embeddings

Al-Subaiey’s Insight:

• Similar accuracy achievable with simpler SVM+TF-IDF
• Complex deep learning not always necessary for email classification
• Trade-off: interpretability (Al-Subaiey has LIME) vs sophistication

Limitation:

• Unspecified dataset hinders reproducibility
• Computational cost likely higher than traditional ML

Dodaj PDF aby wygenerować pełne podsumowanie używając /summarize-paper fang-rcnn-phishing-email-2019