Abstract

The dark web is a hub of illegal activities that includes malware distribution, cyber fraud and extremist operations. It has become a vital source of information for cyber threat intelligence (CTI). The survey covers measurement studies, hidden services, marketplaces, forums, and law-enforcement case investigations, providing a thorough evaluation of dark-web research relevant to CTI. It focuses on recent studies from 2023 to 2025 and investigates the development of dark-web datasets related to cryptomarkets, forums, traffic, authorship, and named-entity recognition. The survey also examines the use of deep learning methods for dark web intelligence extraction, such as multimodal fusion employing attention-based and transformer architectures, detecting criminal intent, classifying encrypted traffic and detecting anomalies. The discussion covers ongoing challenges such as sparse datasets, adversarial behaviour, data distribution shift, scalability and ethical limits. It highlights research gaps and future objectives, emphasizing resilient learning under changing threat situations, real-time multimodal intelligence fusion, resilient data collection, and the incorporation of ethical and legal protections. It can be used by researchers and practitioners to build deep learning-driven CTI platforms based on dark web intelligence.