Knowledge and Information Systems Human-Computer Interaction

Biomedical word sense disambiguation based on capsule gated recurrent unit

Zhang, Chunxiang, Xiang, Bingqi, Gao, Xueyao
Springer Nature · 2026-10-26 · DOI 10.1007/s10115-026-02920-0

Abstract

Words in field of biomedicine usually have ambiguity, so it is very important to understand their meanings correctly for both medical research and clinical practice. In this paper, we propose a method for word sense disambiguation (WSD) of biomedical vocabulary based on capsule gated recurrent unit (GRU). It combines semantic compression ability of capsule neural network with context modeling ability of gated recurrent unit, which can accurately disambiguate polysemous words in biomedical texts. Disambiguation features include word, part of speech, and meanings. Bert is used to vectorize features, GRU is applied to enhance the network’s ability to understand and remember sequences, capsule neural network is adopted to extract spatial hierarchical structure, and linear transformation is used to compute probability distribution of biomedical ambiguous word under each semantic category in classification layer. Semantic category with the highest probability is determined as category of biomedical ambiguous word. We use MSH dataset to optimize the proposed network and testify it. Experimental results show that the proposed method can significantly improve average accuracy of biomedical WSD.

Read at publisher →