Open-Source Intelligence Analysis Method Based on Fine-Tuned Large Models and Knowledge Graphs
Yang Su · 2025
Open-Source Intelligence (OSINT) has become a pivotal component of national security and military intelligence, given that up to 80–90% of intel used by agencies is derived from public sources. However, the massive volume and heterogeneity of open-source data pose challenges for timely, accurate analysis. This paper proposes a concrete methodology for OSINT analysis that synergistically combines a fine-tuned large AI language model with a structured knowledge graph. The fine-tuned model (a transformer-based large language model) is specifically adapted to extract entities, relations, and events from unstructured data and populate a dynamic knowledge graph which enables structured reasoning and querying. We detail the system architecture, including the model’s training process, knowledge graph construction, and the integration framework that allows iterative refinement of intelligence through retrieval-augmented generation. To illustrate the approach, we present a case study in a national security context, where our method builds an intelligence knowledge graph from open sources and uses it to answer complex queries about a hypothetical conflict scenario. Experimental results demonstrate that our fine-tuned model achieves high accuracy on OSINT-specific extraction tasks and that the knowledge graph enhances analytical depth by revealing hidden connections. The proposed OSINT analysis method offers a scalable and actionable framework, improving both the precision and recall of intelligence insights while ensuring they are organized for efficient analyst review.