Optimizing BERT-based Ensembles for Biomedical NER 


Vol. 14,  No. 10, pp. 775-784, Oct.  2025
https://doi.org/10.3745/TKIPS.2025.14.10.775


PDF
  Abstract

In this paper, we presented an approach for named entity recognition(NER) for biomedical corpora using ensemble of BERT-based models to enhance performance of NER system. Specifically, we employed three ensemble strategies for BERT-based models trained on biomedical literature, which are weighted averaging, concatenation, and gating network. The emsemble is conducted at embedding level, in order to capture additional information about specific patterns missed by individual models, hence enhancing expressive power of the language models and mitigating errors by imbalanced data. The candidate BERT-based models were narrowed down by performance evaluations and were optimized via exhaustive search. Our results showed that optimized ensemble models could outperform individual models and provide more reliable predictions.

  Statistics
  Cite this article

[IEEE Style]

K. J. In, Y. S. Won, C. J. Eun, L. K. Cheol, "Optimizing BERT-based Ensembles for Biomedical NER," The Transactions of the Korea Information Processing Society, vol. 14, no. 10, pp. 775-784, 2025. DOI: https://doi.org/10.3745/TKIPS.2025.14.10.775.

[ACM Style]

Kim Jae In, Yoon Seung Won, Cho Jae Eun, and Lee Kyu Cheol. 2025. Optimizing BERT-based Ensembles for Biomedical NER. The Transactions of the Korea Information Processing Society, 14, 10, (2025), 775-784. DOI: https://doi.org/10.3745/TKIPS.2025.14.10.775.

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals