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. A operation be a list in correlation order of the documents whose images are most closely related to q0. [OCRerr] user examines this list and specifies which of the documents in it are relevant and which are not. Since the modification is to be based only on a sample of the relevant documents (assuming that some are missing from the retrieved list associated with q ), the modified request will be formed by adding to 0 the original query, q0, an optimal query vector based on the feedback information. [OCRerr]he resultant[OCRerr]vector (the new query) should thus be a better approximation to the optimal query than q[OCRerr]0, and should, therefore, produce better retrieval when resubmitted. I{ence we seek a relation of the form: q1 = f(q0,[OCRerr],S) where q0 is the original query, [OCRerr] is the subset of the retrieved set which the user deems relevant, and S is the subset of the retrieved set (based on q[OCRerr]0) which the user deems nonrelevant. The form suggested immediately by the &oove is: ni n2 =c(1q0+c(2 L riT12Z (3.8) 1 niL i 1=1 where n1 = n([OCRerr]), n2 = n(S), [OCRerr] =[OCRerr]rž,r2.....r[OCRerr]n1[OCRerr], S =[OCRerr][OCRerr]i' [OCRerr]2.......5n>' where all vectors have been normalized to unit length, and X1 and Oc 2 are arbitrary weighting coefficients.