Skip to main content

Table 1 Comparison of kBSPS in terms of AUC, F1-measure and classification time with other state-of-the-art kernels using the CV and CL evaluation scenarios

From: A fast and effective dependency graph kernel for PPI relation extraction

 

AIMed

BioInfer

IEPA

Kernels

AUC

F 1

 

AUC

F 1

 

AUC

F 1

 
 

CV/CL

CV/CL

time(s)

CV/CL

CV/CL

time(s)

CV/CL

CV/CL

time(s)

SL

83.5/77.5

54.5/42.6

10.8

81.1/74.9

60.0/46.2

24.0

81.1/75.6

69.3/60.4

1.8

SpT

66.1/56.8

27.3/28.6

44.5

74.1/64.2

53.4/43.0

79.8

75.9/54.2

64.7/15.5

0.6

APG

84.6/77.6

56.2/43.8

3.7

81.5/69.6

60.7/39.1

6.2

83.9/82.4

73.1/59.6

0.4

kBSPS

75.1/72.1

44.6/40.3

0.4

75.2/73.3

55.1/47.6

1.5

83.2/81.0

70.5/70.7

< 0.1

  1. Legend: For brevity, only the results on the three largest corpora are shown. Best results are highlighted with bold typesetting (differences under 1 base point are ignored). We reran all experiments since CL results are hardly available. SL - shallow linguistic kernel [1], SpT - spectrum tree kernel [2], APG - all path graph kernel [3].