深度学习在钓鱼网站识别中的应用 -安全工程专业

摘要

随着互联网的快速发展，钓鱼网站已成为网络安全领域的重要威胁之一，其隐蔽性强、传播速度快，传统检测方法在面对复杂多变的攻击手段时逐渐显现出局限性。为此，本研究旨在探索深度学习技术在钓鱼网站识别中的应用潜力，以提升检测效率和准确性。研究选取了多种主流深度学习模型，包括卷积神经网络（CNN）、长短时记忆网络（LSTM）以及Transformer架构，并结合URL特征提取与网页内容分析构建综合识别框架。通过对比实验发现，基于Transformer的模型在处理大规模非结构化数据时表现出显著优势，其准确率较传统机器学习方法提升了约15%，同时具备更强的泛化能力。此外，本研究提出了一种融合多源特征的优化策略，有效弥补了单一特征提取方式的不足，进一步提高了识别性能。研究表明，深度学习技术能够显著增强钓鱼网站检测系统的智能化水平，为实际应用场景提供了可靠的技术支持。该研究的主要贡献在于首次系统性地验证了Transformer架构在钓鱼网站识别任务中的优越性，并提出了适用于复杂场景的多源特征融合方案，为未来相关研究奠定了理论与实践基础。

关键词：钓鱼网站识别；深度学习；Transformer架构；多源特征融合；检测准确性

ABSTRACT

With the rapid development of the Internet, phishing websites have become one of the significant threats in the field of cybersecurity, characterized by strong concealment and rapid dissemination. Traditional detection methods are gradually revealing limitations when confronted with complex and evolving attack techniques. In response to this challenge, this study aims to explore the application potential of deep learning technologies in phishing website identification to enhance both detection efficiency and accuracy. A variety of mainstream deep learning models were selected for investigation, including Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Transformer architectures. These models were integrated into a comprehensive recognition fr amework that combines URL feature extraction with web content analysis. Comparative experiments demonstrated that Transformer-based models exhibit significant advantages in processing large-scale unstructured data, achieving an accuracy improvement of approximately 15% over traditional machine learning methods while demonstrating stronger generalization capabilities. Furthermore, this study proposes an optimized strategy for multi-source feature fusion, effectively addressing the shortcomings of single-feature extraction approaches and further enhancing recognition performance. The findings indicate that deep learning technologies can substantially elevate the intelligence level of phishing website detection systems, providing robust technical support for practical applications. The primary contribution of this research lies in its systematic validation of the superiority of Transformer architectures in phishing website identification tasks and the introduction of a multi-source feature fusion scheme suitable for complex scenarios, thereby establishing a theoretical and practical foundation for future related studies.

Keywords: Phishing Website Identification; Deep Learning; Transformer Architecture; Multi-Source Feature Fusion; Detection Accuracy

摘要 I

ABSTRACT II

第1章绪论 1

1.1 深度学习与钓鱼网站识别的研究背景 1

1.2 国内外研究现状与发展趋势 1

1.3 本文研究方法与技术路线 2

第2章钓鱼网站特征分析与数据准备 3

2.1 钓鱼网站的典型特征提取 3

2.2 数据集构建与标注方法 3

2.3 特征工程在深度学习中的作用 4

第3章深度学习模型设计与优化 5