Skip to main content

Table 5 Intrinsic (left 5 columns) and extrinsic scores (right 4 columns) of different representation models trained on the biomedical corpus

From: Bio-SimVerb and Bio-SimLex: wide-coverage evaluation sets of word similarity in biomedicine

 

UMN-rel(ρ)

UMN-sim(ρ)

MayoSRS(ρ)

Bio-SimVerb(ρ)

Bio-SimLex(ρ)

BC4CHEMD (F-score)

BC2GM (F-score)

AnatEM (F-score)

JNLPBA (F-score)

Attention

0.5248

0.5551

0.6113

0.471

0.7155

79.11

65.91

80.49

62.3

SSG

0.5189

0.552

0.6003

0.4744

0.7181

79.62

67.3

81.3

63.78

SG

0.5767

0.6271

0.5744

0.4638

0.7151

81.37

70.2

81.32

65.16

CBOW

0.5

0.5348

0.5146

0.4367

0.702

78.41

64.05

80.3

61.9

Dependency

0.3934

0.4622

0.3445

0.3978

0.7436

83.69

71.43

82.4

65.01

PubMed-w2v

0.506

0.549

0.5133

0.4376

0.6984

80.71

67.4

81.1

64.86

BioASQ

0.5092

0.5893

0.4729

0.4228

0.6982

56.95

48.86

53.34

50.51

  1. The bolded text implies the best performing models of their kind