Abstract
This paper analyzes existing methods for information retrieval and intelligent classification of data available in closed darknet networks. It considers the features of text data specific to the darknet, such as the presence of specialized slang, jargon, and code expressions. A comparative analysis of traditional search methods and modern approaches based on neural networks, including the BERT model and its adaptation for darknet search, is presented. The challenges associated with the availability and indexing of information are described, and classification methods aimed at improving the accuracy of identifying and filtering illegal content are proposed. The article emphasizes the importance of using machine learning to automate darknet search and monitoring, which can contribute to improving information security.