Reproduce “Classifying spam emails using agglomerative hierarchical clustering and a topic-based approach” on OpenResearch

About this paper

F. Janez-Martino, R. Alaiz-Rodriguez, V. Gonzalez-Castro, E. Fidalgo, E. Alegre

This paper presents a multi-classification system for spam emails that categorizes them into 11 specific cybersecurity topics. The system incorporates techniques to extract text from image-based spam and hidden content, achieving high classification accuracy of over 94% using TF-IDF with Logistic Regression for English spam and TF-IDF with Naïve Bayes for Spanish spam.

Key insights

• Multi-class classification of spam emails into specific cybersecurity topics allows for more nuanced threat intelligence and targeted incident response than traditional binary detection.
• Effective email preprocessing, including Optical Character Recognition (OCR) for image-based spam and detection of hidden text, is essential for robust classification against common evasion tactics.
• Simpler models like TF-IDF combined with Logistic Regression or Naïve Bayes can achieve high accuracy and efficiency in multi-class spam classification, outperforming more complex word embedding approaches like Word2Vec and BERT for the specific datasets used.

Summary by alphaXiv.