Skip to main content
. 2021 Jun 24;25:100422. doi: 10.1016/j.invent.2021.100422

Fig. 3.

Fig. 3

The overall architecture of TransformerRNN. The model contains five components: input embeddings, transformer encoder, BiLSTM, max-pooling layer and classification layer. The symbol ⊕ denotes vector concatenation. The internal architecture of transformer encoder is shown in light green block. More details about our model are provided in the main text.