Table 4 Overview of error categories per model
From: Negation detection in Dutch clinical texts: an evaluation of rule-based and machine learning methods
| Â | Rule-based | biLSTM | RobBERT | |||
|---|---|---|---|---|---|---|
False positives | Â | Â | Â | |||
Ambiguous | 0 | 0% | 14 | 16% | 11 | 13% |
Annotation error | 11 | 3% | 5 | 6% | 13 | 15% |
Minus | 0 | 0% | 8 | 9% | 0 | 0% |
Negation of different term | 119 | 36% | 24 | 27% | 32 | 38% |
Other | 1 | 0% | 13 | 14% | 16 | 19% |
Punctuation | 0 | 0% | 5 | 6% | 1 | 1% |
Scope | 136 | 41% | 6 | 7% | 0 | 0% |
Speculation | 50 | 15% | 10 | 11% | 8 | 9% |
Uncommon negation | 0 | 0% | 5 | 6% | 0 | 0% |
Wrong modality | 14 | 4% | 0 | 0% | 4 | 5% |
Total | 331 | Â | 90 | Â | 85 | Â |
False negatives | Â | Â | Â | |||
Ambiguous | 0 | 0% | 18 | 11% | 8 | 7% |
Annotation error | 20 | 11% | 7 | 4% | 15 | 14% |
Minus | 51 | 27% | 13 | 8% | 23 | 21% |
Negation of different term | 2 | 1% | 0 | 0% | 0 | 0% |
Other | 29 | 15% | 14 | 9% | 15 | 14% |
Punctuation | 13 | 7% | 20 | 12% | 1 | 1% |
Scope | 0 | 0% | 32 | 20% | 8 | 7% |
Speculation | 8 | 4% | 13 | 8% | 20 | 18% |
Uncommon negation | 60 | 32% | 39 | 24% | 17 | 16% |
Wrong modality | 6 | 3% | 6 | 4% | 2 | 2% |
Total | 189 | Â | 162 | Â | 109 | Â |