Enhancing Real Time Situational Awareness with Latent Relationship Discovery

ABSTRACT: The ability to rapidly discover important latent relationships between people, places, events, and locations is critical to maintaining an accurate and informative situational awareness picture. Evidence of such implicit relationships is often distributed across multiple sources and modalities. Existing approaches to discovery of relationships across multiple sources often encounter difficulties aligning disparate data representations, they often make ill-fitting assumptions about the nature of the data, and they frequently lack proven, scalable, online inference procedures. We propose the development of URIEL (Unified Representation and Inference for Exploitation of Links), a software suite that resolves these issues by 1) unifying multiple data source representations into a simple common graph abstraction, and 2) employing a novel application of the flexible generalized Principal Components Analysis (gPCA) framework to infer latent relationships in a principled fashion. Our approach operates online (enabling support to both real-time operations and forensic analysis) and is highly scalable (having been demonstrated on millions of entities in a related context). BENEFIT: The project will create a powerful, flexible, easy-to-use capability for rapidly inferring latent relations from large quantities of disparate multi-source data, which may include arbitrary combinations of structured, semi-structured, and unstructured content. There are immediate applications to multiple pressing Air Force intelligence analysis problems, other intelligence community applications, law enforcement, commercial social network analysis, social media analytics, and business intelligence analysis.