Detection of Social Engineering Attacks Using Natural Language Processing: A Comprehensive Review

Abstract

Social engineering attacks continue to pose significant threats to organizational and individual cybersecurity, exploiting human psychology rather than technical vulnerabilities. With the exponential growth of digital communication channels, attackers have developed increasingly sophisticated techniques to manipulate victims through deceptive messages. Natural Language Processing (NLP) has emerged as a promising approach for detecting and mitigating these attacks by analyzing linguistic patterns, semantic features, and psychological manipulation techniques embedded in malicious communications. This review paper examines the current state of research on applying NLP techniques to detect various forms of social engineering attacks, including phishing emails, spear-phishing campaigns, pretexting, baiting, and social media-based attacks. We systematically analyze the methodologies, datasets, feature extraction techniques, and machine learning models employed in this domain. Furthermore, we discuss the challenges inherent in developing robust detection systems, including adversarial attacks, multilingual detection, contextual understanding, and the evolving nature of social engineering tactics. Finally, we identify promising research directions and opportunities for advancing the field through deep learning architectures, transfer learning, explainable AI, and real-time detection systems.