Table 1
The Results of Concept Matching
| Match Type | Training Set (100 documents) | Test Set (24 documents) | |||
| No. of Matches | Percentage | Distinct Concepts | No. of Matches | Percentage | |
| True positive | 7,227 | 82.6% | 2,268 | 1,298 | 76.3% |
| Redundant UMLS concept | 490 | 5.6% | 209 | 119 | 7.0% |
| Homonym | 481 | 5.5% | 127 | 45 | 2.6% |
| UMLS general concept missing | 158 | 1.8% | 86 | 38 | 2.2% |
| Concept not in UMLS | 127 | 1.5% | 31 | 42 | 2.5% |
| FP, acronym/abbrev | 83 | 0.9% | 51 | 15 | 0.9% |
| FN, variant not in UMLS | 41 | 0.5% | 16 | 44 | 2.6% |
| FN, inferable by ctx/expert | 38 | 0.4% | 12 | 31 | 1.8% |
| FN, acronym/abbreviation/elision | 29 | 0.3% | 6 | 37 | 2.2% |
| Concept not useful for indexing | 25 | 0.3% | 7 | 6 | 0.4% |
| Too many non-stop-words | 25 | 0.3% | 25 | 7 | 0.4% |
| FN, spelling/grammar error | 8 | 0.1% | 8 | 0 | 0.0% |
| FN/FP, proper name | 10 | 0.1% | 10 | 19 | 1.1% |
| FP, spelling/grammar error | 3 | 0.0% | 3 | 0 | 0.0% |
| totals: | 8,745 | 2,859 | 1,701 | ||
-
NOTES: The three columns indicate category of match, the number of matches for each category, and the number of distinct concepts matched. FN indicates false negative; FP, false positive. The number of negated concepts in the test set was 110.
