- Authorities Identified the Owner of the Hacker forum BreachForum’s Primary Fetish - August 17, 2023
- Rival Markets Place Offers to Win Over Former ASAP Vendors and Customers - August 17, 2023
- How Hackers Use the Dark Web? - August 17, 2023
South Korean researchers have successfully trained DarkBERT, a neural network designed to comprehend the slang used by participants on the Darknet. Tom’s Hardware reports on this groundbreaking achievement, citing the technical neural documentation. The researchers from the Institute of Advanced Technology in South Korea trained DarkBERT using a vast array of Darknet text data, consisting of approximately 6 million web pages, and leveraging the RoBERTa language model.
Developing DarkBERT: Advancing Understanding of Darknet Slang and Limitations of Neural Networks
One notable aspect of this development is the distinct nature of Darknet slang, which possesses specific characteristics enabling attackers to evade detection. For instance, Russian-speaking hackers often employ Anglicisms and even modify syntax to camouflage their public exchanges on forums, thereby concealing their activities.
Through the training of the neural network, scientists have made significant strides in detecting confidential data leaks on the internet. DarkBERT is capable of discerning whether published data represents a new leak or a duplication of previously disclosed information. Moreover, the neural network has acquired the ability to identify advertisements for the illicit sale of psychotropic substances by analyzing the presence of specific keywords. This innovative solution has the potential to streamline the efforts of law enforcement agencies in apprehending offenders involved in such activities.
It is important to note that DarkBERT, despite its accomplishments, does have limitations. Currently, the technology only supports the English language, limiting its applicability in multilingual Darknet contexts. Additionally, the learning process is still a manual endeavor due to variations in site layouts, necessitating the compilation of data arrays for neural network training.
Despite the widespread enthusiasm surrounding neural networks, it is crucial to recognize that the technology is still a work in progress. A case in point is the performance of ChatGPT and Sage neural networks, which received a “C” grade on a history exam at the Ural Federal University. Mikhail Kiselyov, an associate professor of history in UrFU, asserts that the chatbots’ algorithms simulate independent reasoning, resulting in the fabrication of facts and references to non-existent scholarly works. These challenges underscore the ongoing journey to refine and enhance the capabilities of neural networks.