Table 1.

Hyperparameter settings in our model.

Hyperparameter	Optimum value	Search range
Learning rate	5×10-5	{1×10-3, 1×10-4, ..., 1×10-6}
Batch size	64	{16, 32, 64, 128, 256}
Sentence vector size	300	{100, 200, 300, 400, 500}
Dropout rate	0.5	{0.1, 0.2, 0.3, ..., 0.8}
Down-sampling rate	0a	{0, 0.1, ..., 1}

^aThe optimum setting had no down-sampling.