Interactive Information Retrieval Systems with Minimalist Representation

Almost any information you might want is becoming available on-line. The problem is how to find what you need. One strategy to improve access to existing information sources, is intelligent information agents-an approach based on extensive representation and inference. Another alternative is to simply concentrate on better information organization and indexing. Our systems use a form of conceptual indexing sensitive to users' task-specific information needs. We aim for minimalist representation, coding only select aspects of stored items. Rather than supporting reliable automated inference, the primary purpose of our representations is to provide sufficient discrimination and guidance to a user for a given domain and task. This paper argues, using case studies, that minimal representations can make strong contributions to the usefulness and usability of interactive information systems, while minimizing knowledge engineering effort. We demonstrate this approach in several broad spectrum applications including video retrieval and advisory systems.