Biomedical Named Entity Recognition Based on Residual Network and Global Context Mechanism

Biomedical Named Entity Recognition (BNER) is a fundamental and important task in biomedical text mining, aiming at automatically identifying and classifying biomedical entities. To improve the performance of BNER method, a Residual Network (ResNet) and Global Context Mechanism (GCM) based BNER method is proposed. Which uses Transformer for layer normalization in the feature extraction layer and BiLSTM to obtain the contextual information of sentences; residual network is used to alleviate the problem of vanishing gradient in the feature fusion layer, and the global context mechanism is used to integrate the sentences that existed in the first cell of the BiLSTM and the last cell in BiLSTM to integrate the sentence representation in each cell. We conducted extensive experiments on five benchmark BNER datasets, Experimental results have shown that our model performs best on the JNLPBA and LINNEAUS datasets.