Skip to main content

Table 3 Effect of different knowledge node representations base on knowledge graph on the CDR and CHR test sets (%)

From: Document-level medical relation extraction via edge-oriented graph neural network based on document structure and external knowledge

Dataset

Method

Overall

Intra

Inter

P

R

F1

F1

F1

CDR

SEoG

64.5

75.5

69.6

73.4

59.9

SKEoG (TransE)

65.7

75.7

70.4

74.0

60.5

SKEoG (RESCAL)

67.7

70.4

69.0

73.0

58.4

SKEoG (GAT)

64.7

72.5

68.4

72.6

57.6

CHR

SEoG

88.3

92.6

90.4

93.1

84.4

SKEoG (TransE)

89.9

91.4

90.6

93.2

85.0

SKEoG (RESCAL)

90.6

92.1

91.4

93.6

86.4

SKEoG (GAT)

87.5

93.6

90.5

93.1

84.6

  1. Bold highlight the highest result on a given dataset in our experiments