Building a Discourse-Argument Hybrid System for Vietnamese Why-Question Answering

Table 2

Research works on why-question answering.

Author	Year	Methodology	Dataset	Result
Verberne	2006–2010	IR + RST relation classification	Selected 186 English why-questions on INEX corpus	MRR@150 = 0.34
Higashinaka and Isozaki	2008	IR + causal relation classification using SVM	Dataset developed in Japanese	MRR@20 = 0.339
Oh et. al.	2013	IR + causal extraction using CRF	WhySet, dataset developed in Japanese	P@1 = 41.8%
	2016	IR + causal extraction using CRF, augmented by adding more training data	WhySet	P@1 = 50%
	2017	IR + causal extraction using CRF, answer selection using CNN network	WhySet	P@1 = 54%
	2019	IR + GAN-like network (GAN–generative adversarial network)	WhySet	P@1 = 54.8%
			Quasar-T (https://github.com/bdhingra/quasar)	EM = 43.2%; F1 = 49.7%
			SearchQA	EM = 59.6%; F1 = 65.3%
			TriviaQA	EM = 49.6%; F1 = 54.8%