Scholarlink

<p>Biomedical Named Entity Recognition (BioNER) extracts entities such as diseases, drugs, and genes from biomedical texts, which are often dense in domain-specific terms and complex semantics. Here, we present FRKAN-BioNER, a model designed to improve the efficiency of data mining in the biomedical field and support the development of precision medicine knowledge graphs. FRKAN-BioNER integrates BioBERT (Bidirectional Encoder Representations from Transformers for Biomedical Text Mining) with the Fourier Kolmogorov-Arnold Network (FourierKAN). The KAN architecture addresses limitations in traditional neural networks, improving model expressiveness and trainability. The model achieved F1-score of 84.80%, 93.12%, 90.02%, 82.10%, 87.90%, 83.14%, 78.58%, 89.93%, and 90.87% across nine public datasets. These results demonstrate that FRKAN-BioNER outperforms several prior state-of-the-art methods. Furthermore, its innovative architecture may hold potential for improving the efficiency of BioNER-relevant clinical text processing and could help accelerate knowledge mining from large-scale biomedical literature.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

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

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月27日

查看全文
  • Li Yelin,
  • Wu Yan,
  • Xie Xiaojun,
  • Zhu Jihong,
  • Guan Lixin,
  • Li Mengshan

摘要

Biomedical Named Entity Recognition (BioNER) extracts entities such as diseases, drugs, and genes from biomedical texts, which are often dense in domain-specific terms and complex semantics. Here, we present FRKAN-BioNER, a model designed to improve the efficiency of data mining in the biomedical field and support the development of precision medicine knowledge graphs. FRKAN-BioNER integrates BioBERT (Bidirectional Encoder Representations from Transformers for Biomedical Text Mining) with the Fourier Kolmogorov-Arnold Network (FourierKAN). The KAN architecture addresses limitations in traditional neural networks, improving model expressiveness and trainability. The model achieved F1-score of 84.80%, 93.12%, 90.02%, 82.10%, 87.90%, 83.14%, 78.58%, 89.93%, and 90.87% across nine public datasets. These results demonstrate that FRKAN-BioNER outperforms several prior state-of-the-art methods. Furthermore, its innovative architecture may hold potential for improving the efficiency of BioNER-relevant clinical text processing and could help accelerate knowledge mining from large-scale biomedical literature.