Fig. 3.

Comparison between MTT and state-of-the-art methods on the Novel Ebola and Novel H1N1 datasets over different combinations of negative training and testing sets. MTT is significantly better than Doc2vec on the Novel Ebola dataset (a), while on the Novel H1N1 dataset (b), the reverse holds true. MTT is statistically better than Denovo on both datasets. For the Generalized model, we can only have results up to the negative training rate of 2 because, for larger negative training rates, the model took days to finish one run