Skip to main content

Table 7 Error analysis (number of facts in brackets)

From: KnowLife: a versatile approach for constructing a large knowledge graph for biomedical sciences

  

Percentage based on text genre

Percentage

Cause of error

Encyclopaedic

Scientific

Social

  

sources

sources

sources

8.16% (62)

Preprocessing

38.71% (24)

3.23% (2)

58.06% (36)

27.24% (207)

Entity Recognition

13.04% (27)

45.41% (94)

41.55% (86)

32.11% (244)

Entity Disambiguation

12.30% (30)

26.23% (64)

61.48% (150)

1.97% (15)

Coreferencing

13.33% (2)

13.33% (2)

73.33% (11)

13.68% (104)

Nonexistent Relation

23.08% (24)

29.81% (31)

47.12% (49)

9.21% (70)

Pattern Relation Duality

24.29% (17)

27.14% (19)

48.57% (34)

3.29% (25)

Swapped left and right-hand entity

28.00% (7)

24.00% (6)

48.00% (12)

3.03% (23)

Negation

17.39% (4)

21.74% (5)

60.87% (14)

1.32% (10)

Factually Wrong

40.00% (4)

10.00% (1)

50.00% (5)