Fourier Kolmogorov-Arnold Network integrated into BioBERT-based model for Biomedical Named Entity Recognition.

Area of Science:

• Biomedical informatics
• Natural Language Processing
• Machine Learning

Background:

• Biomedical texts contain complex semantics and domain-specific terms, posing challenges for data mining.
• Biomedical Named Entity Recognition (BioNER) is crucial for extracting key information like diseases, drugs, and genes.
• Existing methods may have limitations in expressiveness and trainability for complex biomedical data.

Purpose of the Study:

• To introduce FRKAN-BioNER, a novel model for efficient Biomedical Named Entity Recognition (BioNER).
• To enhance data mining capabilities within the biomedical domain.
• To support the development of precision medicine knowledge graphs.

Main Methods:

• Integration of BioBERT (Bidirectional Encoder Representations from Transformers for Biomedical Text Mining) with the Fourier Kolmogorov-Arnold Network (FourierKAN).
• Leveraging the KAN architecture to improve model expressiveness and trainability compared to traditional neural networks.

Main Results:

• Achieved high F1-scores across nine public datasets, with scores ranging from 78.58% to 93.12%.
• Demonstrated superior performance compared to several existing state-of-the-art BioNER methods.
• FRKAN-BioNER achieved an average F1-score of 84.80% across datasets.

Conclusions:

• FRKAN-BioNER offers an effective solution for improving BioNER.
• The model's innovative architecture shows potential for efficient clinical text processing.
• This approach can accelerate knowledge discovery from large-scale biomedical literature for precision medicine.