Table 2.
Recall and precision of temporal extraction algorithm
Gold standard | True positives | False positives | False negatives | Recall | Precision | F measure | |
Past dates | 373 | 349 | 10 | 14 | 0.93 | 0.97 | 0.95 |
Specified | 297 | 285 | 3 | 7 | 0.96 | 0.99 | 0.97 |
Relative | 78 | 64 | 7 | 7 | 0.82 | 0.90 | 0.86 |
Future dates | 123 | 101 | 8 | 14 | 0.82 | 0.93 | 0.87 |
Specified | 30 | 19 | 2 | 9 | 0.63 | 0.90 | 0.75 |
Relative | 93 | 82 | 6 | 5 | 0.88 | 0.93 | 0.91 |
Recurring dates | 13 | 9 | 0 | 4 | 0.69 | 1.00 | 0.82 |
Present dates | 29 | 29 | 0 | 0 | 1.00 | 1.00 | 1.00 |
Totals | 538 | 488 | 25* | 32 | 0.91 | 0.95 | 0.93 |
True positives represent dates correctly identified and assigned to individual colonoscopy concepts. Specified dates are any explicit date formats (eg, ‘3/06’, ‘early 1990s’). Relative dates include any reference that requires a calculation (eg, ‘5 years ago’, ‘last Monday’).
Seven false positives were identified from sentences without a colonoscopy date (n=670).