A Multitask Deep Learning Framework for DNER

Table 3

The parameters for our experiments.

Layer	Hyperparameter	Value
CNN	Window size	3
	Number of filters	30
LSTM	State size	200
	Initial state	0.0
	Peepholes	No
Dropout	Dropout rate	0.5
	Batch size	10
	Initial learning rate	0.015
	Gradient clipping	5.0
	Decay rate	0.05
	Labeling schema	BIO
	ELMo dim	1024