Available under a Creative Commons Attribution Non-Commercial Share Alike 4.0 International Licence
1.2 COMPUTER AND INFORMATION SCIENCE, Computer Sciences, Information Science
Recent studies have shown that contextualized word embeddings outperform other types of embeddings on a variety of tasks. However, there is little research done to evaluate their effectiveness in the biomedical domain under multi-task settings. We derive the contextualized word embeddings from the Flair framework and apply them to the task of biomedical NER on 5 benchmark datasets, yielding major improvements over the baseline and achieving competitive results over the current best systems. We analyze the sources of these improvements, reporting model performances over different combinations of word embeddings, and fine-tuning and casing modes.
Akhtyamova L., Cardiff J. (2020) LM-Based Word Embeddings Improve Biomedical Named Entity Recognition: A Detailed Analysis. In: Rojas I., Valenzuela O., Rojas F., Herrera L., Ortuño F. (eds) Bioinformatics and Biomedical Engineering. IWBBIO 2020. Lecture Notes in Computer Science, vol 12108. Springer, Cham. DOI:10.1007/978-3-030-45385-5_56