Skip to main content
. 2022 Feb 24;62(5):1207–1213. doi: 10.1021/acs.jcim.1c01199

Figure 4.

Figure 4

Structure of the NER model created by using the Bert vectors as contextualized embeddings (a) which are then fed into a bidirectional LSTM (BiLSTM). These contextual embeddings are supplemented by character embeddings in the variant (b). The shaded parts of the model were trained, while the unshaded parts were left frozen.