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

BMC Bioinformatics

Table 3 Performances of the proposed model and other baselines (average of five iterations) with Double Deep Q Network (DDQN) algorithm as policy learning method

Model	Diagnosis success rate	Dialogue length	AMR (%)	AMR2 (%)	Disease classifier accuracy (DC) (%)
Flat policy [10]	0.3370	5.43	2.82	1.32	/
HRL (Liao et al., 2020) [12]	0.4784	8.57	13.80	29.24	48.40
PR-SIDDA with only AM	0.4906	8.92	11.34	29.21	49.08
PR-SIDDA with only RM	0.4984	8.58	14.96	35.76	49.01
PR-SIDDA	0.4820	7.74	15.32	33.10	49.22
A-SIDDA with only AM	0.4737	8.73	11.40	27.50	47.31
A-SIDDA with only RM	0.4996	8.27	14.78	33.44	49.04
A-SIDDA	0.5201	8.34	16.04	40.54	52.52

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

ISSN: 1471-2105