Skip to main content

Table 5 Performance comparison of the proposed and baseline methods on the MedConSult

From: Composition-driven symptom phrase recognition for Chinese medical consultation corpora

\(\varepsilon\) Method Macro/% Micro/%
Precision Recall \(F_1\) Precision Recall \(F_1\)
0.6 ComD 41.02 37.92 39.41 39.01 21.84 28.00
ComD-Character 36.23 33.37 34.74 34.13 13.62 19.47
BERT-CRF 47.17 16.57 24.52 47.16 15.32 23.13
BiLSTM-CRF 36.50 29.58 32.68 34.01 8.43 13.51
BDMM-based 8.23 12.15 9.81 5.92 8.58 7.00
Dictionary-based 13.52 7.72 9.83 13.76 7.72 9.89
0.7 ComD 33.53 30.08 31.71 31.11 17.42 22.33
ComD-Character 28.36 25.76 27.00 26.30 10.49 15.00
BERT-CRF 37.56 13.05 19.38 37.44 12.03 18.21
BiLSTM-CRF 32.00 25.80 28.57 26.18 7.51 11.67
BDMM-based 7.14 10.46 8.49 5.10 7.37 6.02
Dictionary-based 9.51 5.69 7.12 9.48 5.38 6.86
0.8 ComD 27.45 24.26 25.75 25.23 14.12 18.11
ComD-Character 21.81 19.44 20.55 19.78 7.89 11.28
BERT-CRF 27.52 9.48 14.10 27.45 8.82 13.35
BiLSTM-CRF 19.25 14.94 16.82 14.99 3.61 5.82
BDMM-based 6.17 8.77 7.25 4.22 6.15 5.00
Dictionary-based 8.65 5.14 6.44 8.72 4.94 6.31
1.0 ComD 25.05 22.62 23.77 23.37 13.08 16.78
ComD-Character 19.42 17.26 18.28 17.61 7.03 10.04
BERT-CRF 26.56 9.11 13.57 26.55 8.53 12.91
BiLSTM-CRF 15.00 12.52 13.65 12.59 3.03 4.89
BDMM-based 5.94 8.54 7.01 4.10 5.98 4.86
Dictionary-based 8.65 5.13 6.44 8.72 4.94 6.31
  1. The best result with bold font for each parameter/model/characteristic