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. 1. Read an initial query vector q[OCRerr] (in integer format), convert it to a unit vector, and store it in the array Q(I), I = 1,i\. ([OCRerr] is the dimension of the index language vector space.) 2. Read in the set of relevant document vectors r. j = 1 ,n1, convert them to unit vectors, and store them in the array R(i,j), I = 1 ,N; J = 1 ,[OCRerr] 1 [OCRerr]. Read in the set of nonrelevant document vectors 5., 3. j = 1 ,n2, convert them to unit vectors, and store them in the array s(i,j), I = 1,N; J = 2 (Note that since the dimension of the index space was N = 511.for[OCRerr]hetthesaurusuused, and since a document vector typically has a[OCRerr]out [OCRerr]5 nonzero components, the program actually handled the vectors in a condensed format;) 4. Form a new query vector represented by the arr[OCRerr]y: Q1(I) = N1N2Q(I) + N1 N2 N2[OCRerr]1.R(I[OCRerr]J) - N1 [OCRerr] s(i,j) J=1 5. Normalize to unit length: [OCRerr] ZNQl (1)2 L 1=1 j Program Steps for Producing Relevance Feedback [OCRerr]odified Queries Table 5.1