SP500215 Okapi at TREC-2 chapter S. Robertson S. Walker S. Jones M. Hancock-Beaulieu M. Gatford National Institute of Standards and Technology D. K. Harman Table 5: Manual searches with feedback % of tops where Run AveP PS P30 P100 RP Rd AveP > median Official (BM15) 0.232 0.492 0.468 0.400 0.297 0.591 40 Re-run (BM11) 0.247 0.480 0.477 0.411 0.315 0.607 48 Database: disks 1 and 2. Topics: 101-150 Table 6: Some results from query modification #fbk Term % of tops where docs source k3 Terms AveP P5 P30 P100 RP Rd AveP > median 50 TCNDdoc 8 all 0.369 0.660 0.591 0.487 0.408 0.754 92 30 TCNDdoc 8 all 0.368 0.668 0.585 0.486 0.407 0.749 88 10 TCNDdoc 8 all 0.363 0.668 0.584 0.485 0.400 0.748 88 50 TCNDdoc 9 all 0.360 0.624 0.573 0.482 0.399 0.748 88 50 TCNDdoc 9 top 40 0.354 0.620 0.573 0.478 0.394 0.741 86 50 TCNDdoc 9 top 30 0.353 0.632 0.567 0.480 0.395 0.742 88 30 TCNDdoc 8 top 30 0.360 0.676 0.577 0.479 0.402 0.741 88 30 TCNDdoc 8 top 20 0.348 0.636 0.571 0.474 0.392 0.734 82 30 TCNDdoc 8 top 10 0.318 0.604 0.537 0.449 0.366 0.702 78 50 TCND 8 all 0.364 0.636 0.573 0.487 0.406 0.755 92 30 TCND 8 all 0.362 0.644 0.573 0.484 0.408 0.749 90 10 TCND 8 all 0.363 0.640 0.574 0.481 0.406 0.754 88 30 TCND 0 all 0.334 0.652 0.559 0.458 0.374 0.711 80 30 TCNDdoc 0 all 0.310 0.645 0.546 0.448 0.359 0.675 66 Initial feedback run for comparison (top row of Table 2) None TCND large all 0.360 0.652 0.569 0.479 0.401 0.754 92 Retrospective run using all known relevant documents to reweight the topic terms Variable TCND 0 all 0.371 0.708 0.600 0.497 0.408 0.758 92 Database: disks 1 and 2. Topics: 101-150 34