Automatic Methods for Linking Texts to a Knowledge Base Using Neural Networks
Multiphase Systems · 30.09.2026 · 21 (3) · 137-145
DOI: 10.21662/mfs2026.3.021 ↗Kurz gefasst
Fachartikel in der Zeitschrift Multiphase Systems, online erschienen am 30.09.2026. Verfasst von Igor L Kuzmin.
Abstract (Originalsprache)
Linking biomedical terms to medical knowledge base concepts is an important step in medical text processing and is used in decision support systems, medical searches, and scientific publication analysis. Despite significant progress in the development of neural network methods, existing solutions still face challenges in processing multilingual data, the high variability of medical terminology, and the limited coverage of Russian-language terms in the UMLS knowledge base. The goal of this work is to develop a method for linking biomedical terms to UMLS concepts that ensures high quality for both English and Russian. To address this problem, we developed a method that includes preparing a UMLS knowledge base, constructing a FAISS index for quickly finding the closest candidates, using the SapBERT and BERGAMOT models to generate term embeddings, automatically translating Russian-language mentions, re-ranking the identified candidates based on semantic and lexical features, and aggregating the results at the UMLS concept level. An experimental evaluation was conducted on the official BioNNE-L dataset, containing English, Russian, and multilingual tracks. The proposed method demonstrated competitive results on all three tracks: Russian, English, and multilingual. An error analysis revealed that the main reasons for the poor quality were insufficient coverage of Russian-language terminology in UMLS, homonymy of medical terms, and difficulty processing abbreviations. The obtained results confirm the effectiveness of the proposed method for linking biomedical terms to UMLS concepts. The developedapproach can be used in the creation of medical information systems and decision support systems, and also serve as a basis for further research in the field of multilingual biomedical entity linking and medical terminology normalization.
Zitieren
APA
Kuzmin, I. L. (2026). Automatic Methods for Linking Texts to a Knowledge Base Using Neural Networks. Multiphase Systems, 21(3), 137-145. https://doi.org/10.21662/mfs2026.3.021
Vancouver
Kuzmin IL. Automatic Methods for Linking Texts to a Knowledge Base Using Neural Networks. Multiphase Systems. 2026;21(3):137-145. doi:10.21662/mfs2026.3.021
BibTeX
@article{Kuzmin2026,
author = {Kuzmin, Igor L},
title = {Automatic Methods for Linking Texts to a Knowledge Base Using Neural Networks},
journal = {Multiphase Systems},
year = {2026},
volume = {21},
number = {3},
pages = {137--145},
doi = {10.21662/mfs2026.3.021}
}
EUVIMED