Table 10 Token-wise precision (Pr), recall (Re) and F1-score (F1) results for medication information extraction per class and per model, including entity-wise F1-score in brackets and the micro average F1-score in the last row.
From: A distributable German clinical corpus containing cardiovascular clinical routine doctor’s letters
Class type | CRF | BERT | Count token (entities) in CARDIO:DE100 | ||||
|---|---|---|---|---|---|---|---|
Pr | Re | F1 | Pr | Re | F1 | ||
ActiveIng | 0.84 | 0.83 | 0.83 (0.60) | 0.80 | 0.91 | 0.85 (0.86) | 1,596 (1,479) |
Drug | 0.80 | 0.75 | 0.77 (0.77) | 0.81 | 0.87 | 0.84 (0.81) | 532 (414) |
Duration | 0.80 | 0.73 | 0.77 (0.60) | 0.78 | 0.89 | 0.83 (0.59) | 1,514 (294) |
Form | 0.47 | 0.33 | 0.39 (0.41) | 0.57 | 0.71 | 0.63 (0.60) | 24 (20) |
Frequency | 0.97 | 0.97 | 0.96 (0.94) | 0.96 | 0.98 | 0.97 (0.94) | 6,471 (1,341) |
Strength | 0.93 | 0.96 | 0.94 (0.92) | 0.93 | 0.97 | 0.95 (0.93) | 2,692 (1,341) |
Micro avg. | 0.92 | 0.91 | 0.92 (0.87) | 0.90 | 0.95 | 0.93 (0.88) | 12,829 (4,889) |
