Course-of-Action Forecasting

Intelligence contains adversary course-of-action plans embedded into unrelated information sources. This information in ASCII format often contains nothing more than subject and title, thus requiring tagging. Source material is almost always dated, or time stamped. Likewise, source data may be spatially enabled with a geographic location and if it is not, it can be linked via a gazetteer. Exploration of spatial data mining and temporal data mining has been carried out mostly independently in the knowledge discovery research community although the object-oriented machine learning algorithms can certainly be adapted to both. Here we propose to develop a geospatial pattern recognition and tracking system based on our spatiotemporal similarity metrics and learning agent expertise using the information extraction platform of Inxight Software. Our system will extract geospatial entities, concepts and facts (e.g. people, places, things, events, relationships) from the unstructured intelligence data, classify the extracted data into taxonomies of topics for more targeted information access, enable the intelligence analyst visualize and explore the geospatial data with graphical and text-based browsing, concept and full-text search. In essence, we model the information extraction platform outputs with a belief network that tracks the identified geospatial patterns and provides COA forecast.