Skip to main content

Table 2 Test accuracies (in percent) on 20 words from our word-specific models and the two baselines

From: Biomedical word sense disambiguation with bidirectional long short-term memory and attention-based neural networks

Words

Basic NN

Sum NN

Cct-V NN

Cct-T NN

Cct-T Atten

Cct-T NN wp

Baseline 1

Baseline 2

AA

100.00

100.00

100.00

100.00

97.20

100.00

96.00

98.99

Astragalus

100.00

100.00

100.00

100.00

100.00

100.00

100.00

97.47

CDR

100.00

100.00

100.00

100.00

96.40

100.00

97.00

100.00

Cilia

96.15

92.31

96.15

96.15

93.80

92.00

82.00

94.87

CNS

91.18

94.12

91.18

97.06

97.20

100.00

98.00

98.48

CP

96.30

96.30

96.30

96.30

100.00

100.00

97.00

98.32

dC

97.22

100.00

97.22

97.22

100.00

100.00

98.00

98.48

EMS

97.14

97.14

97.14

97.14

100.00

100.00

98.00

100.00

ERUPTION

97.14

100.00

94.29

100.00

100.00

100.00

100.00

100.00

FAS

100.00

100.00

100.00

100.00

100.00

100.00

100.00

99.49

Ganglion

91.67

91.67

94.44

88.89

88.90

91.18

90.00

93.43

HCl

97.22

97.22

97.22

97.22

100.00

100.00

100.00

100.00

INDO

100.00

100.00

100.00

100.00

100.00

100.00

87.00

99.18

lymphogranulomatosis

94.44

100.00

94.44

100.00

95.80

100.00

83.00

83.33

MCC

87.50

100.00

100.00

100.00

100.00

100.00

97.00

100.00

PAC

100.00

100.00

100.00

90.91

100.00

100.00

94.00

100.00

Phosphorus

86.11

86.11

83.33

83.33

91.70

94.44

78.00

83.84

Phosphorylase

86.67

90.00

86.67

90.00

78.10

90.00

52.00

87.35

TMP

100.00

100.00

100.00

100.00

100.00

100.00

81.00

98.00

TNT

100.00

100.00

100.00

100.00

97.20

100.00

98.00

99.49

Average

95.94

97.24

96.42

96.71

96.82

98.38

91.30

96.54

  1. Baseline 1 is the RCNN model by Festag in [14], and baseline 2 is the Naive Bayes Model by Jimeno-Yepes et al. in [25]. Cct-T Atten means the word-specific attention model with structure (iii), and Cct-T NN wp means the word-specific BiLSTM neural network model with structure (iii) trained on whole-paragraph inputs. All the word specific models here (deep network and attention) are equipped with layer \(\mathcal {C}\)