Abstract
The advent of decentralized and anonymized web infrastructures and the Dark Web specifically, has severely changed the situation with cyber crime. The Dark Web supports the black market through anonymity networks like Tor, I2P, and Freenet: ransomware-as-a-service, malware trading, financial fraud, and coordinated cyberattacks. These activities challenge cybersecurity agencies with serious concerns because traditional methods of key-based or even shallow NLP don't detect the contextual, multilingual, and adversarial language that is common in communications on the Dark Web. This chapter explores how Transformer-based Natural Language Processing (NLP) models overcome them. It examines the language peculiarities of Dark Web materials such as obfuscation, code-mixing, and language peculiarities of domain and emphasizes the limitations of the traditional methods of surveillance. The paper introduces an end-to-end threat intelligence pipeline based on the Transformer with the implementation of multiple stages such as Dark Web crawling, adversarial text preprocessing etc.

This publication has 12 references indexed in Scilit: