Skip to main content

Table 8 Overall performance of CBOW and Skip-gram according to the voting system: Number of unique UniProtKB entries and number of term pairs for protein/gene names that are involved in Experiment I, II, and combined (i.e. merging Experiment I and II)

From: Deep learning meets ontologies: experiments to anchor the cardiovascular disease ontology in the biomedical literature

 

Voting system

Experiment

Model

Number of terms pairs

Number of UniProtKB entries

Number

FTV

Number FTV for top three

Number TV

(%)

I

CBOW

1020

64

31

21

43 (67%)

II

CBOW

816

63

29

21

49 (78%)

I and II

combined

CBOW

1836

79

47

37

64 (81%)

I

Skip-gram

1020

64

49

37

57 (89%)

II

Skip-gram

816

63

56

51

60 (95%)

I and II

combined

Skip-gram

1836

79

71

63

77 (97%)

  1. According to the voting system, for each model the last three columns show: the number of full term variants among the top twelve ranked candidate terms for the UniProtKB entries (Number FTV column); the number of full term variants among the top three ranked candidate terms for the UniProtKB entries (Number FTV for top three); and the number and % of term variants (i.e. FTV and/or PTV) among the top twelve ranked candidate terms for the UniProtKB entries (Number TV column)