Table 4 Medical event detection accuracy of our sequence labelling models on the test set
From: An approach for medical event detection in Chinese clinical notes of electronic health records
| Â | Model | Acc. |
|---|---|---|
Text classification models | bi-LSTM(1 layer) | 82.4% |
| Â | bi-LSTM(2 layers) | 84.2% |
| Â | bi-LSTM(3 layers) | 83.1% |
| Â | text-CNN | 85.8% |
Sequence labelling models | bi-LSTM(2 layers) feature extractor + bi-LSTM decoder | 87.3% |
| Â | bi-LSTM(2 layers) feature extractor + smoothed Viterbi decoder | 87.4% |
| Â | bi-LSTM(2 layers) feature extractor + CRF decoder | 92.1% |
| Â | CNNs feature extractor + bi-LSTM decoder | 89.9% |
| Â | CNNs feature extractor + smoothed Viterbi decoder | 90.7% |
| Â | CNNs feature extractor + CRF decoder | 92.6% |