This work proposes a novel, classificatory
analysis based relevance feedback framework based on a
user-centric model of information need that is independent
of any particular retrieval paradigm. The model of the user
need is based on the principle that a complete
representation of the user need is contained in an
exhaustive user classification of the collection. This model
provides a conceptually appealing basis for relevance
feedback; each successive iteration of relevance feedback
can be treated as a classification that becomes a closer
approximation of the user's information need. The system
iteratively achieves a better understanding of the user's
information need, gradually converging to a satisfactory set
of results. The framework is based on Rough Set Theory,
which is explicitly designed to deal with classificatory
analysis incorporating uncertainty and approximation.