Exploring Security Vulnerabilities of Deep Learning Models by Adversarial Attacks

Table 5

Accuracy changes when the num of replaced words increases (Yelp review dataset).

Num of replaced words	0	10	20	30	40	50
LSTM	0.9569	0.9296	0.8906	0.7812	0.6640	0.6171
BiLSTM	0.9564	0.8984	0.8906	0.7578	0.6562	0.5546