IRS13 Appendix A: Recall-Precision Tables appendix Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. A-i Appendix A Recall-Precision Tables Introduction To supplement the numerous precision versus recall graphs given in this report the tables of the original average data are presented in this appendix for 79 retrieval results. Each result is labelled with the collection and request set, type of correlation function, length of document, dictionary, and other variables involved. Ten recall ("R") and precision ([OCRerr][OCRerr]ptt) pairs are given for each result using a cut-off at ten recall levels, together with the "().uasi-Cranfield" interpolation method (Fig. 16a, Section II) and the third method of "left end extra- polation" (Fig. 20, Section II). In the individual requests any cases of tied ranks are resolved by using the document identification numbers in numerical order to give a unique rank position to each such document. Following the recall versus precision results are the four single figure evaluation measures: Rank Recall (Rank Rec.), Log Precision (Log Prec.), Normalized Recall (Nor. Rec.) and Normalized Precision (Nor. Prec.). These are derived using a tied rank breaking method in which all tied documents are assigned the same average rank number. The results given correspond exactly with the presentation of normalized recall and precision and precision versus recall curves made in sections III, V, VI, VII, and X.