Centro de Documentação da PJ
Analítico de Periódico
CD 358
Utilisation of Large Language Models (LLMs) in OSINT-based cyberterrorism detection on social media [Recurso eletrónico] / Ala Berzinji, Mazyar Farhad Abdalmajid
International Journal of Cyber Criminology, Vol. 18, n. 1 (January-June 2024), p. 210-223
Ficheiro de 480 KB em formato PDF.
CIBERTERRORISMO, REDE SOCIAL, REDE NEURONAL ARTIFICIAL, INTELIGÊNCIA ARTIFICIAL
This paper explores a novel approach to identifying extremist content on social media platforms, with a particular focus on Twitter, in the context of countering cyberterrorism. Large language models (LLMs) have been effective in many natural language processing tasks, and the performance is usually limited by the scope of their training data. To deal with the limits, the research suggests an innovative method integrating LLM-based techniques, particularly the LLaMA3 model, with open-source intelligence (OSINT) strategies for enhancing detecting extremist content on Twitter. The key innovation of this approach lies in using a Retrieval-Augmented Generation (RAG) model powered by LLaMA3, incorporating OSINT data to provide a broader context for Twitter posts. The incorporation of external sources of information, news articles, government reports, and academic publications, the RAG model suggests a more robust framework to identify extremist rhetoric. This allows us to detect subtle linguistic markers and potential signs of radicalisation otherwise being overlooked. The paper analyses a sample of Twitter data using the proposed framework, which demonstrates its effectiveness in the identification of extremist content, even from a single tweet. According to the results, this integrated methodology significantly enhances the detection of nuanced extremist discourse on social media platforms.
