ISR11 A Modified Two-Level Search Algorithm Using Request Clustering chapter V. R. Lesser Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. VII-19 0.6 12 c=0.2 n=3 in steps in steps of 3 of 0.1 o.6 12 [OCRerr] Kc,n c=0.2 n=3 0.6 12 Kc,n P = [OCRerr] P(qi.,c,n) T c=0.2 fl=3 j=1 3 0.6 12 [OCRerr] Kc,n c=0.2 n=3 12 Kc,n [OCRerr] R(qi.,c,n) n=3 j=1 3 0.6 12 [OCRerr] Kc,n c=0.2 n=3 M(qi .,c,n) 3 0.6 R c=0.2 T Kc,n j=l ** The values of [OCRerr] P[OCRerr] and R represent avera[OCRerr]e values for the criteria I T over the entire range of user needs. This provides a measure of search effectiveness for a givcn search scheme, and a given set of categories based on a test collection of queries. The values of M (co,no), P (co,no), R (co,no) provide the same type of measure of search efficiency, except that these measures are related to a particular user need (e.g. high recall or high precision, etc.) * could be calculated in the following viay: 0.6 12 _ = [OCRerr] M (c,n) c=0.2 n=3 20 However for a limited set of queries this method of calculating is not valid since Kc, n for c, n large will be_very small, (i.e. covering not many cases), and therefore the value of M (c, n) can fluctuate arbitrarily for such a small sample, so that its value is not a good indicator of the search effectiveness, and thus should not be given an equal weight in the averaging procedure. ** The averaging technique used to calculate these criteria is similar to the procedure used to calculate ranked recall. [6]