Word Sense Disambiguation of Medical Abbreviations Using Semantic Types Classification

Abstract

Word Sense Disambiguation (WSD) is a crucial task in Natural Language Processing (NLP), typically in specialized domains like biomedicine, where the same abbreviation can have multiple meanings. The integration of semantic knowledge and knowledge graphs into WSD approaches has been essential for improving the clarity and correctness of medical texts. This paper presents an approach to semantic type based WSD of biomedical abbreviations using fine-tuned Bidirectional Encoder Representations from Transformers (BERT) models. The proposed approach leverages the contextual information from surrounding words and their associated semantic types, extracted from the Unified Medical Language System (UMLS), to disambiguate the sense of abbreviations. Two state-of-the-art BERT models, BioBERT and ELECTRA, are fine-tuned on a large dataset of biomedical text and evaluated their performance on the MeDAL test set manually curated for semantic type based WSD. The results demonstrate significant improvements in accuracy and F1-score compared to baseline methods, showcasing the effectiveness of incorporating semantic type information into resolving the task of WSD in medical text.