Model-guided information discovery for intelligence analysis

Intelligence analysis can be aided and guided by models of the analysts' interests and priorities. This paper describes our approach to analyst modeling as part of the Ant CAFÉ project, in which analyst models are used to guide the searching behavior of a swarm of intelligent agents. Structural elements of our analyst model include concepts and relations, both of which help to capture the analyst's current interest and concerns. In addition, the concepts and relationships have associated scalar parameters to provide a quantitative measure of the user's level of interest. We have developed algorithms for dynamically adapting the weights and evolving the elements of the model itself. To evaluate these algorithms we have built an Analyst Modeling Environment workbench. We have tested our approach on this workbench using traces generated by human analysts, and have demonstrated improvements over current state of the art search engines.