Skip to main content

Table 1 Accuracy (%) MA and NE recognition

From: Performance analysis of CRF-based learning for processing WoT application requests expressed in natural language

Size of training data

MA recognition

NE recognition

Test set

Test set

200

7000

200

7000

1000

38.8

41.4

28.1

26.9

3000

42.8

45.8

29.1

29.9

5000

49.3

51.1

29.4

31.3

7000

42.0

51.3

29.4

30.2

9000

44.5

51.6

32.8

32.7

  1. MA recognition is relatively more promising in identifying triggers and actions