IRS13 Search Matching Functions chapter E. M. Keen Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. iii-i8 EVALUATION [OCRerr]U[OCRerr][OCRerr]ER [OCRerr] [OCRerr]R[OCRerr]ENTpGE* MEASURE USED OF INDIVIDUAL REQUESTS COLLECTION INPUT AND DICTIONARY TO DETERMINE COSINE OVERLAP BOTIJ [OCRerr] MERIT SUPERIOR SUPERIOR EQU([OCRerr]L t IRE-3 Abstract, Stem Normed Recall 20 58.8% l[OCRerr] [OCRerr]i.2% 0 Normed Precision 21 61.8% 13 38e2% 0 3[OCRerr] Requests Abstract, Thesaurus Normed Recall 26 76.5% 8 23.5% 0 -3 Normed Precision 22 6[OCRerr].7% 12 35.3% 0 CRMT-l Abstract, Stem Normed Recall 28 70.0% 12 30.0% 2 Normed Precision 314 82.9% 7 17.1% 1 42 Abstract, Thesaurus Normed Recall 35 83.3% 7 16.7% 0 Requests -3 Normed Precision 36 85.7% 6 14.3% C) Text, Stem Normed Recall 24 70.6% 10 29.4% 1 Normed Precision 23 67.6% 11 22.4% 1 ADI Text, Thesaurus-l Normed Recall 26 74.3% 9 25.7% 0 Normed Precision 28 8o.o% 7 20.0% 0 35 Requests Abstract, Stem Normed Recall 17 5O.0[OCRerr] 17 5o.0% 1 Normed Precision 18 52.9% 16 47.1% 1 Abstract, Thesaurus Normed Recall 23 67.6% 11 32.4% 1 -1 Normed Precision 28 8o.o% 7 20.0% 0 * Percentages do not include cases where both options have equal merit. Comparison of individual request merit givin[OCRerr] the numbers of requests forming cosine and overlap with percentages on 8 options from three collections, according to merit assigned by normalized recall and precision. Figure 8.