One of the most successful techniques developed for document retrieval is relevance feedback. The effectiveness of this technique can be attributed to the fact that it provides a simple method for interactively acquiring knowledge from the user that is not available from the query. In systems that use knowledge-based and natural language processing techniques, the ability to interactively acquire domain and linguistic knowledge from users becomes crucial. This is because, in any real application, this knowledge will be incomplete or missing entirely. This research involves experiments explicitly designed to test the hypothesis hat users of an information systems can interactively provide knowledge that will improve retrieval effectiveness. The experiments will address the acquisition of three types of knowledge: (1) Query knowledge such as the relative importance of words and phrase, (2) Domain knowledge such as concepts that are related to the query concepts, and (3) Linguistic knowledge that allows the system to understand compound nouns. The interfaces that are used to acquire this knowledge will be constructed using the document retrieval system built at the University of Massachusetts. The structure of the linguistic knowledge that is acquired will be based on the representation language REST that is part of the ADRENAL project at the University of Massachusetts.
