The senses in which computers and humans may be said to "understand" themselves and each other in the environment of computer systems for document retrieval are discussed. While the extent to which computers understand is still at a rather low level, many recent attempts at retrieval system performance can be seen to involve an attempt to achieve greater understanding. Three paradigms of retrieval methodology - deep semantic, statistical, and "smart Boolean" -are contrasted for their approaches from a knowledge-based perspective. A detailed summary of one approach in the smart Boolean framework - the CONIT intermediary retrieval assistance system - is given with respect to its attempts at providing understanding to the computer and the human. It is shown how CONIT incorporates in its workings knowledge of the retrieval systems and their databases, the user's problem, effective search heuristics, the dynamics of the search itself, the effectiveness of search results, and search strategy modification techniques. Particular attention is focussed on newly designed techniques for estimating precision, for ranking documents by estimated relevance, and for search strategy modification based on user'relevance feedback.
展开▼
