SP500215 Appendix A: TREC-2 Results Appendix National Institute of Standards and Technology D. K. Harman suits - HNC, Inc. L Summary Statistics Run Number HNCad2-full, feedback Num of Queries 50 1 1 Total number of documents over all queries 0.8< Retrieved: 50000 Relevant: 10785 Rel[OCRerr]ret: 6911 0.6 Recall-Precision Curve [OCRerr]ecall Level Averages Recall Precision 0.00 0.7811 0.10 0.5920 0.20 0.5262 0.30 0.4429 0.40 0.3574 0.50 0.2767 0.60 0.2059 0.70 0.1250 0.80 0.0713 0.90 0.0218 1.00 0.0006 `rage precision over all vant docs -interpolated 0.2882 [OCRerr]t Level Averages ut Recall 31 [OCRerr] 0.4132 [OCRerr]20 0.5207 2 [OCRerr]0 0.5790 )60 0.6233 1 )80 0.6485 0 .00 0.6643 L20 0.6676 -1 L40 0.6676 L60 0.6676 -2 L80 0.6676 -3 ?00 0.6676 Document Level Averages Precision At 5 docs 0.5560 At 10 docs 0.5680 At 15 docs 0.5507 At 20 docs 0.5470 At 30 docs 0.5513 At 100 docs 0.4784 At 200 docs 0.3714 At 500 docs 0.2243 At 1000 docs 0.1382 R-Precision (precision after R (where R is relevant docu- docs retrieved the number of ments)) Exact 0.3575 NORMAL DEVIATE - Fallout-Recall .001 0.02 0.16 0.5 0.84 0.98 0.C I I I I -~7-- -3 -2 -1 0 1 2 3 Precision 0.4 0.2 0 1 0.8 0.6 I[OCRerr]999 0.4[OCRerr] 0.98 0.2 0.84 0.5 0 Recall I I 0 0.2 0.4 0.6 0.8 1 Recall Fallout-Recall Curve I I 0 0.05 0.1 0.15 0.2 0.16 Fallout x 100 0.02 0.001