By all accounts, the Web is humanity's largest and fastest growing repository of digital information. Many collections of information are Internet-accessible, and most will provide a searchable Web interface. While some collections have a broad array of materials, trends show an explosion in the number of specialized collections with narrow but very deep content. Thus a principle challenge facing users will be the selection of Web information sources capable of answering their query. In a physical library, users rely on a reference librarian to help point them at the correct resource, but while human librarians are becoming increasingly sophisticated in their use of the Web, they are only part of the solution. We need more powerful automatic reference tools to help people efficiently retrieve high quality information from the Web. Typically, reference librarians are not specialists in the topic of inquiry (e.g., computational fluid dynamics) but they are expert at identifying relevant resources (e.g., The International Journal of Fluid Dynamics) and at appropriate strategies for obtaining the necessary information. The central objective of this proposal is to create software agents that posses reference intelligence --- a limited understanding of complex technical topics, but a very sophisticated understanding of how and where to find high-quality information on the World Wide Web.
