ISR10 Search Request Formulation chapter Joseph John Rocchio Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. user, there is no certainty that an ideal search req[OCRerr]est (under some fixed index transformation) in fact exists. [OCRerr]elevance, the relation between the query and the subset D is a function of the user1s infor- mation needs and his interpretation of the text of reference documents; while the retrieval ordering produced by the system is a function of the query and document index representations In the case where there is no ideal[OCRerr]request corresponding to a given subset D[OCRerr], one might say that the index transformation is deficient from the point of view of the particular user (who specified i[OCRerr]), since it does not allow distincti[OCRerr]ns equivalent to those he can make In general one [OCRerr]must a[OCRerr]sume that this wilL be the norm[OCRerr]rather than the exception,. since the indexing[OCRerr]process is designed to reduce rather than preserve information. It is therefore useful to define an unambiguous, optimal search request as a function of DR, D, and implicitly, therefore, of the index transformation, such.:thatfor every non-empty unique `subset D of D, this optimal search request both exists and is R unique. An o[OCRerr]timal search request index image corresponding to a given subset D of a collection of'document images produced by the R index transformation T is that request image q which maximizes the difference between the mean of its correlations with the relevant documents (members of DR) and the mean of its correlations with the nonrelevant documents (members of D' not in D ). R The subjective relevance relation which specifies a subset D corresponding to each input query induQes an effective partition R