A Multichannel Biomedical Named Entity Recognition Model Based on Multitask Learning and Contextualized Word Representations

<div>Multichannel BiGRU-CRF architecture. It consists of 5 parts: input layer, embedding layer, BiGRU, merge layer, and CRF layer. The red, green, and blue lines, respectively, represent the channels after the concatenate operation of two representations, and the yellow line represents the channel after the concatenate operation of all three representations. <i>H</i>0-6, respectively, denotes the bidirectional single-channel GRU. Each channel is independent, which avoids information redundancy.</div>

Wireless Communications and Mobile Computing

fig1

Figure 1

Figure 1: A Multichannel Biomedical Named Entity Recognition Model Based on Multitask Learning and Contextualized Word Representations 