基于文本语义认知操控分析的钓鱼邮件检测方法
A Phishing Email Detection Method Based on Cemantic Cognitive Control Analysis
黄瀚娆 1武斌2
作者信息
- 1. 北京邮电大学网络空间安全学院，北京 100876
- 2. 北京邮电大学网络空间安全学院灾备与数据安全中心，北京 100876
- 折叠
摘要
随着网络攻击手段的不断演进，钓鱼邮件正逐渐从简单的广撒网模式转向利用复杂的社会工程学原理进行心理操控。现有的检测方案多依赖于邮件的表层结构特征或通用的文本语义表示，难以有效捕捉深层的欺诈意图与认知陷阱，导致在面对高伪装性攻击时存在漏报率高、泛化能力弱的问题。为此，本文设计了一种基于认知操控意图引导的深度学习模型。该模型在DeBERTa预训练语言模型的基础上，创新性地引入了认知引导交叉注意力机制，通过提取细粒度的认知操控特征动态引导语义编码器聚焦于关键的欺诈线索；同时，结合多特征门控融合网络自适应地平衡文本语义信息与认知操控意图特征的权重。实验表明，该模型在混合数据集上的F1值达到98.72%，召回率达到98.88%，在准确性与鲁棒性方面优于部分现有的先进基线模型，验证了其在复杂文本环境下识别社会工程学攻击的有效性。
Abstract
With the continuous evolution of cyber attack methods, phishing emails are gradually shifting from a simple broad-sweeping model to psychological manipulation through the utilization of complex social engineering principles. The existing detection schemes mostly rely on the surface structural features of emails or general text semantic representations, making it difficult to effectively capture the deep fraudulent intentions and cognitive traps, resulting in high false negative rates and weak generalization capabilities when facing highly deceptive attacks. Therefore, this paper designs a deep learning model based on cognitive manipulation intention guidance. This model is based on the DeBERTa pre-trained language model and innovatively introduces a cognitive-guided cross-attention mechanism, extracting fine-grained cognitive manipulation features to dynamically guide the semantic encoder to focus on key fraudulent clues.At the same time, it combines a multi-feature gated fusion network to adaptively balance the weights of text semantic information and cognitive manipulation intention features. Experiments show that the model achieves an F1 value of 98.72% and a recall rate of 98.88% on a mixed dataset, outperforming some existing advanced baseline models in terms of accuracy and robustness, verifying its effectiveness in identifying social engineering attacks in complex text environments.关键词
数据安全与计算机安全/钓鱼邮件检测/注意力机制/深度学习/心理认知Key words
data security and computer security/phishing email detection/attention mechanism/deep learning/psychological cognition引用本文复制引用
黄瀚娆,武斌.基于文本语义认知操控分析的钓鱼邮件检测方法[EB/OL].(2026-02-12)[2026-02-14].http://www.paper.edu.cn/releasepaper/content/202602-74.学科分类
计算技术、计算机技术
评论