Scholarlink

Phishing attacks pose a serious concern wherein original websites are replicated to steal sensitive information. Despite numerous strides made by traditional machine learning techniques in countering phishing threats, there persist some challenges especially on feature extraction and computational efficiency aspects. This research introduces a new method using Graph Neural Network (GNN) together with advanced techniques for creating features to detect phishing URLs. This embedded Graph Feature Network (EFGN) model not only targets the structural data of the URL but also interrelations between various elements found on different sites essentially forming a foundation for web pages appearance and function. The GNN architecture enables effective learning of dependency patterns in URL datasets while feature embedding techniques such as word embeddings and one-hot encoding enhance the representation of URL components. The proposed model integrates a dual-pathway architecture where it can capture both relational features by GNN and individual URL components processed at a parallel embedding layer. Experiments were conducted on EGFN using data sets such as PhishTank and OpenPhish and the findings indicate that the proposed model surpasses others in terms of accuracy, precision, recall and F1-score.

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

Enhancing Phishing URL Detection with Graph Neural Networks and Feature Embedding Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Manika Nanda,
  • Mala Saraswat,
  • Pankaj Kumar Sharma

摘要

Phishing attacks pose a serious concern wherein original websites are replicated to steal sensitive information. Despite numerous strides made by traditional machine learning techniques in countering phishing threats, there persist some challenges especially on feature extraction and computational efficiency aspects. This research introduces a new method using Graph Neural Network (GNN) together with advanced techniques for creating features to detect phishing URLs. This embedded Graph Feature Network (EFGN) model not only targets the structural data of the URL but also interrelations between various elements found on different sites essentially forming a foundation for web pages appearance and function. The GNN architecture enables effective learning of dependency patterns in URL datasets while feature embedding techniques such as word embeddings and one-hot encoding enhance the representation of URL components. The proposed model integrates a dual-pathway architecture where it can capture both relational features by GNN and individual URL components processed at a parallel embedding layer. Experiments were conducted on EGFN using data sets such as PhishTank and OpenPhish and the findings indicate that the proposed model surpasses others in terms of accuracy, precision, recall and F1-score.