A Deep Learning Filter that Blocks Phishing Campaigns Using Intelligent English Text Recognition Methods.

Area of Science:

• Cybersecurity
• Machine Learning
• Natural Language Processing

Background:

• Sophisticated cybercrime increasingly relies on specialized phishing campaigns.
• Identifying and classifying phishing patterns across diverse campaigns remains a significant challenge due to limited generalized labeling schemes.
• Existing methods often use feature engineering or neural networks (LSTM, CNN), but can lose crucial semantic information, impacting performance.

Purpose of the Study:

• To propose an enhanced method for detecting phishing campaigns by extending word embeddings.
• To improve the semantic understanding of text data for more accurate phishing campaign classification.
• To develop a novel approach that integrates semantic similarity with existing neural network architectures.

Main Methods:

• Extending word embeddings with word vectors representing semantic similarity to phishing campaign template tags.
• Calculating embedded keywords based on semantic subfields derived from automatic keyword extraction.
• Utilizing sequential Kalman filters to combine general word embeddings with similarity-based vectors.
• Powering neural architectures like LSTM and CNN with the enhanced embeddings for prediction.

Main Results:

• The proposed approach demonstrates remarkable results in phishing campaign detection.
• The method effectively captures and utilizes semantic information lost in traditional models.
• Experimental evaluation using a data indicator confirms the approach's effectiveness and superiority.

Conclusions:

• The novel method of extending word embeddings with semantic similarity significantly enhances phishing campaign detection.
• This approach offers a robust solution to the challenge of classifying diverse phishing campaigns.
• The findings reinforce the state-of-the-art in cybersecurity threat identification using advanced machine learning techniques.