Table 4.

Classification accuracies of tweets, using (i) bag-of-words features (BOW), (ii) proposed features (PRO). Diagonal entries are for in-domain classification, while the non-diagonal entries are for cross-domain classification. Values in the bracket represent standard deviations in case of in-domain accuracies

Train set	Test set
	Ebola		MERS
	BOW	PRO	BOW	PRO
Ebola	84.78% (0.05)	84.02% (0.06)	65.69%	76.15%
MERS	66.19%	74.72%	88.26%(0.07)	81.05% (0.03)

In-domain classification results are represented by italic entries. For each train-test pair, the accuracy of better performing system has been boldfaced