Model-Guided Mining and Discovery

In this talk I will describe the research done at Sarnoff developing adaptive models to tailor the behavior of search and knowledge discovery systems to the needs of individual users. I will describe our development of dynamically adaptive models that capture user intent and entity behavior, and can dynamically adapt by both expanding model content (with the aid of an ontology) and by refining measures of user interest (based on implicit and explicit user feedback). I will describe previous and current work in the context of several applications and present experimental evaluation results. I will also describe how these models are being used to develop an anomaly detection system for surveillance applications, and outline our recent work in threat assessment algorithms. The talk will conclude with a short description of the commercial applications of our work.