SCAFFOLD – Synthesizing Course of Action Frames From Observations and Link Derivation

Predictive analytics tools, such as Insight, rely on example behavior models to feed their internal forecasting and inferencing components.  These behavior models represent potential courses of action (COAs) for targets of interest.  Currently, they are produced from scratch by human experts—a costly and inefficient use of specialized human resources that bottlenecks the analytic process. To address this inefficiency, Stottler Henke proposes to develop the SCAFFOLD system to automatically construct the core “who/what/when/where” of these behavior models from reports of real events.  SCAFFOLD will advance the state of the art by addressing three shortcomings of automated event extraction.  First, the SCAFFOLD system will build COAs from co-referring events and entities in different sources—even when they are represented differently among those sources.  Second, SCAFFOLD will correlate multi-source observations to “fill in the gaps” between geospatially referenced observations.  Finally, SCAFFOLD will infer implicit relationships to identify which other events may influence an extracted COA. This effort leverages our past work and experience in the areas of event and entity extraction, natural language processing, and predictive modeling. Phase I development of a limited prototype will provide a solid foundation for the complete implementation of SCAFFOLD in Phase II, and its eventual commercialization.