ISR10 Evaluation of Document Retrieval Systems 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 5-5 4 =[OCRerr]-Fr[OCRerr]onret rieval an[OCRerr] [OCRerr]nrelevance[OCRerr] (5.4) P. From this joint probability distribution a number of conditional probabilities can be defined as follows (followin[OCRerr] Swets, reference 1): n1/(n1+n3) = Fr [OCRerr]Retrieval/[OCRerr]elevance2 = [OCRerr] (5.5) the conditional probability of a l[OCRerr]hit!I; n2/(n2+n4) = Pr £[OCRerr]etrieval/[OCRerr]onrelevance} = [OCRerr] (5.6) the conditional probability of a [OCRerr]1false drop11; n3/(n1+n3) = Fr [OCRerr][OCRerr]onretrieval%[OCRerr]elevanc[OCRerr] *= (5.7) the conditional probability of a 1?missII; n4/(n2+n4) = Pr [OCRerr][OCRerr]onr.etri::eval/[OCRerr]onrelevance[OCRerr] = [OCRerr] (5.8)[OCRerr] the conditional probability of a 1lcorrect rejection'.'. 6 [OCRerr]ote that n1/(n1+n) is the recall ratio as defined by Cleverdon 3 while n1/(n1+n2), the conditional probability of relevance given retrieval, is his precision (also called relevance) ratio. [OCRerr]he Bernoulli-like model assumed above is in many respec?s A Bernoulli model assumes repeated: independent trials in which there are only two possible outcomes for each trial and the probabilities of eachoutc'6me remain constant throughout the experiment.