ISR10 Scientific Report No. ISR-10 Information Storage and Retrieval 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. 5-21 1 for 1:< i [OCRerr] n0 p*(i) = I [OCRerr] [OCRerr]O for n0< [OCRerr]< [OCRerr] and where n0 = n(IH), i.e., the number of relevant documents to the query under consideration; N n(D), the number of documents in the reference collection; and 1 = the rank index induced on D. [OCRerr]he function r(i) is viewed as the number of relevant documents having rank order less than or equal to i divided by the total number of relevant documents. Thus, it[OCRerr]is Cleverdon1s recall as a function of the order induced on D by a retrieval operation. Clearly, r*(i) is the recall functidn which pertains when the retrieval operation produces an ideal ordering on D. Similarly, p(i) is the number of relevant documents having rank order less than or equal to i divided by i, with p*(i) defined for the case when all members of D have a rank index less than every member of D . Hence R q for each query q,.[OCRerr]r*(i) definesa:desired (or objective) recall q function, and p*(i) defines a desired precision function. q Since it has been assumed that H induces only an ordering on D, as opposed to a metric, these functions are strictly defined only for discrete values of[OCRerr]the rank index i. As it is intended to extend these fu[OCRerr]ctions to a continuous independent variable, that is, to