Skip to main content

Table 2 Performance of the proposed model and other baselines (average of five iterations) with Deep Q Network (DQN) algorithm as policy learning method

From: Symptoms are known by their companies: towards association guided disease diagnosis assistant

Model

Diagnosis success rate

Dialogue length

AMR (%)

AMR2 (%)

Disease classifier accuracy (DC) (%)

SVM-ex

0.3210

/

/

/

32.10

REFUEL Peng et al. [15]

0.3470

4.56

/

16.10

/

KR-DS Xu et al. [31]

0.3570

6.24

/

38.80

/

GAMP Xia et al. [14]

0.2670

1.36

/

7.70

/

Flat policy

0.3420

5.34

2.41

1.26

/

HRL Liao et al. [12]

0.5040

12.95

10.49

29.41

49.80

PR-SIDDA with only AM

0.5226

12.44

9.38

36.85

52.86

PR-SIDDA with only RM

0.5182

9.14

13.90

37.38

52.14

PR-SIDDA

0.5162

9.94

13.52

40.84

51.42

A-SIDDA with only AM

0.5260

11.69

9.75

33.70

53.53

A-SIDDA with only RM

0.5378

11.34

12.34

42.66

54.18

A-SIDDA

0.5576

11.79

12.68

43.82

56.16

  1. AM and RM refer to the association module and recommendation module, respectively