Semantic Targeting using Analyst Role, Topics and Entity Recognition (STARTER)

Typical search technology ignores the intended use of documentsmissing the opportunity to tailor prioritization and anticipate supplemental materials. Commercial campaigns employ semantic targetingtailoring content to a users predicted preferenceswhich we could incorporate into analytical workflows. What is needed is a sophisticated approach to incorporating the user as part of the data model, leveraging the data, behaviors and interests of the user, and prior analysts. Because analyst turnover is rampant, embedding legacy knowledge into the data model is critical. We will build on the semantic reasoning foundation we established in Phase I to provide a prototype search capability that automatically adapts search using analysts roles. We will learn how role-specific individuals select and navigate content, follow leads, and identify relevant data in complex, multi-step tasking to accomplish this vision. To minimize the training data burden, we rely on minimally-invasive approaches that discover the roles, semantic content, and semantic connections. We will leverage and extend work in probabilistic topic modeling to include the role information and complex operational tasking. We will work closely with PMW 150 to prepare for transition of STARTER to operational Naval users.