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. 3-10 of R and the mean of its correlations with the members of S. Second, it can easily be shown from the definition of the vector dot product that C is maximized by the vector q'= q1 subject to the condition that 0 the components of q' be nonnegative. [OCRerr]he components of q? are given by: 0 q0[OCRerr] 0. = L 0 if q >0 0. = 1,N) if q <0 C) (3.7) ilence, under the assuniptions made, an unambignous optimal (for the criteria stated) query image exists corresponding to any non- empty subset of D. Further, the equation 3.5 provides an effective means of generating such a query from knowledge of the relevant subset [OCRerr] In the evaluation of information retrieval systems and in particular in the evaluation of the indexing function of such systems, this `formulation of an optimal search request provides the ability to isolate' the effects of ind'exing from variances due to request formulation. An optimal search request measures the ability of the index transformation to, differentiate a particular set of documents from all the others of a collection. In an evaluation situation, where one assumes prior kn[OCRerr]wledge of the document subset relevant to each test query, the retrieval performance of the optimal query corresponding to the relevant subset provides a direct measure ,of the ability of the system to extract from the index representations of documents the same kind of `information the user can' extract from, the natural lang[OCRerr]age.