Biomedical Named Entity Recognition Based on BioBERT-ABiLSTM-CRF Modeling

The realm of biomedical named entity recognition (BioNER) poses a considerable challenge in the domain of natural language processing (NLP). In recent years, deep learning-based methods have emerged as a prominent approach within this field. The majority of deep learning models focus on the extraction of features from words and characters during the preprocessing stage. However, this approach neglects the semantic information embedded within the context of a word, which can result in the inability to accurately recognise the multiple meanings of a word. Additionally, the accuracy of named entity recognition methods that utilise static word vectors to characterise semantics is often limited. In order to address these issues, this paper proposes a model combining the pre-trained language model BioBERT with the classical BiLSTM-CRF model and introducing an attention mechanism. This model is applied to the BioNER task by first generating dynamic word vectors based on contextual information. Information is extracted through preprocessing with the BioBERT model, then inputted into the BiLSTM model to extract textual features. This is followed by the Attention Mechanism Layer, which captures information related to the current entity. Finally, an optimal sequence of labels is outputted through the CRF Layer, which outputs the predicted entity categories. The model achieved <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{F1}$</tex> values of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${8 6. 5 2 \%,} {8 7. 5 7 \%}$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${8 3. 8 3 \%}$</tex> for the BC2GM, BC5CDR, and NCBI-disease datasets, respectively. The experimental results demonstrate that the proposed model effectively enhances the accuracy of BioNER.