BioByGANS: biomedical named entity recognition by fusing contextual and syntactic features through graph attention...

Area of Science:

• Biomedical text mining
• Natural Language Processing
• Bioinformatics

Background:

• Biomedical named entity recognition (BioNER) is crucial for extracting knowledge from unstructured biomedical texts.
• Current BioNER models often use deep neural networks but underutilize sentence syntactic features.
• Integrating semantic and syntactic features is essential for improving BioNER performance.

Purpose of the Study:

• To propose a novel BioNER model that effectively integrates semantic and syntactic features.
• To address the limitations of existing models in utilizing sentence dependencies and topology.
• To improve the accuracy and efficiency of biomedical knowledge extraction.

Main Methods:

• Developed BioByGANS (BioBERT/SpaCy-Graph Attention Network-Softmax) model.
• Formulated BioNER as a node classification problem using graph representations of sentences.
• Utilized BioBERT for contextual features and SpaCy for syntactic features (POS, dependencies, topology).
• Employed a Graph Attention Network (GAT) to fuse contextual and syntactic information.

Main Results:

• BioByGANS achieved state-of-the-art performance on 8 benchmark datasets.
• Achieved high F1-scores across datasets, including BC2GM (85.15%), JNLPBA (78.16%), and BC4CHEMD (92.97%).
• Demonstrated superior performance compared to existing BioNER methods.

Conclusions:

• The proposed BioByGANS model significantly improves BioNER performance.
• Formulating BioNER as a node classification task with graph attention networks is effective.
• Integrating syntactic features enhances the model's ability to capture linguistic nuances.