Table 3.

Tested value ranges included in the hyper-parameter search for our multi-modal machine learning architecture.

Parameter	Range
LSTM layer dimension	64, 128, 256, 512
Number of LSTM layers	1, 2, 3, 4
CNN filters	14, 16, 18
CNN kernel sizes	[1,4,7]
CNN pool sizes	3, 5, 7
Dense layer dimension	8, 16, 32, 64, 128, 256, 512
Dropout	0.1, 0.3, 0.5
Batch size	20, 40, 60
Learning rate	10−1, 10−2, 10−3, 10−4, 10−5
Random seeds	13, 22, 42, 66, 78
Threshold	0.3, 0.5, 0.7

Threshold only applies to relation detection.