OSPREY: Multi-INT Contextual Understanding

VSI with support from STR proposes OSPREY, a flexible and robust ontological framework to perceive and understand the content and context of multi-INT and OSINT data. OSPREY will automatically extract ontological entities and relationships using a pre-trained language model and automatically arrange the entities and relationships into a flexible and dynamic knowledge graph without the need for an explicit hand-crafted ontology. This knowledge graph is not only machine readable but also human readable, allowing subject matter experts the opportunity to inspect the information, verify its accuracy, and update as needed. These Knowledge Graphs can then be used in conjunction with an LLM to enhance machine understanding and contextual reasoning. Finally, OSPREY encourages human-machine collaboration by allowing an analyst to pose natural-language queries that are answered based on OSPREYÆs understanding of the content and context of the processed data, and visualize the results in a User Interface that includes visual analytics and drilldowns to highlight various aspects of the media evidence.