Skip to main content
. 2020 May 4;8(5):e17637. doi: 10.2196/17637

Table 3.

Comparison of method performance on the Chinese electronic medical record (CEMR) dataset.

Method P value (%) R value (%) F1 score (%)
Conditional random field (CRF) 88.57 68.43 77.21
CNNa+BiLSTMb+CRF 81.51 76.92 79.15
Lattice long short-term memory (LSTM) 88.60 74.48 80.93
Bidirectional Encoder Representations from Transformers (BERT) 83.73 78.76 81.17
Multi-level representation learning for entity recognition (multi-level ER) 85.21 79.23 82.11

aCNN: convolutional neural network.

bBiLSTM: bidirectional long short-term memory.