Reproduce “Large Language Models for Explainable Threat Intelligence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tiago Dinis, Miguel Correia, Roger Tavares
The RAGRecon system offers explainable Cyber Threat Intelligence (CTI) to cybersecurity personnel by integrating Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) and presenting reasoning through dynamic Knowledge Graphs (KGs). This approach consistently achieves high faithfulness scores (over 0.8) and efficient context utilization, making LLM-generated CTI more transparent and actionable for both conventional and blockchain security threats.
Key insights
- RAG systems effectively enhance LLMs for CTI by providing real-time, domain-specific context, thus improving precision and significantly reducing hallucinations.
- Visualizing LLM reasoning through Knowledge Graphs fosters transparency and trust, allowing cybersecurity analysts to understand the basis of AI-generated intelligence.
- LLM self-evaluation can be a highly accurate method for assessing faithfulness and context relevance in RAG systems, providing reliable feedback on generative performance.
- Small and medium-sized LLMs can struggle with the precise instruction following needed for consistent structured output, such as Knowledge Graph JSON generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.