KoRASA: Pipeline Optimization for Open-Source Korean Natural Language Understanding Framework Based on Deep Learning

Table 7

Performance comparison of intent classification.

Model	Intent classification				Entity extraction
	Train		Test		Train		Test
	Acc	F1	Acc	F1	Acc	F1	Acc	F1
Keyword	0.975	0.982	0.782	0.786	0.993	0.989	0.957	0.908
Fallback	1.000	1.000	0.980	0.981	0.993	0.989	0.959	0.905
DIET-Base	1.000	1.000	0.982	0.980	0.993	0.989	0.962	0.910
KoRASA	1.000	1.000	0.982	0.984	0.993	0.989	0.974	0.947