Deep Learning-Based Methods for Sentiment Analysis on Nepali COVID-19-Related Tweets

Table 4

The architecture of three CNNs proposed in our work.

Layer
	(n, s)	Output shape	(n, s)	Output shape	(n, s)	Output shape
Input	—	(17, 1)	—	(3, 1)	—	(300, 1)
Conv1D + Relu	(32, 3)	(15, 32)	(8, 2)	(2, 8)	(32, 3)	(298, 32)
Conv1D + Relu	(16, 3)	(13, 16)	(8, 2)	(1, 8)	(16, 3)	(296, 16)
Flatten + Dropout (0.2)	—	208	—	8	—	4736
Dense + Dropout (0.2)	—	128	—	6	—	128
Dense	—	64	—	4	—	64
Softmax	—	3	—	3	—	3

Note that (n, s) denotes the number of filters and filter size for the corresponding CNN model.