We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A Multi-pass Sieve for Clinical Concept Normalization.
- Authors
Yuxia Wang; Hur, Brian; Verspoor, Karin; Baldwin, Timothy
- Abstract
Clinical concept normalization involves linking entity mentions in clinical narratives to their corresponding concepts in standardized medical terminologies. It can be used to determine the specific meaning of a mention, facilitating effective use and exchange of clinical information, and to support semantic cross-compatibility of texts. We present a rule-based multipass sieve approach incorporating both exact and approximate matching based on dictionaries, and experiment with back-translation as a means of data augmentation. The dictionaries are built from the UMLS Metathesaurus as well as MCN corpus training data. Additionally, we train a multi-class baseline based on BERT. Our multi-pass sieve approach achieves an accuracy of 82.0% on the MCN corpus, the highest for any rule-based method. A hybrid method combining these two achieves a slightly higher accuracy of 82.3%.
- Subjects
SIEVES; MEDICAL terminology; HEALTH information exchanges; CONCEPTS; SYSTEMATIZED Nomenclature of Medicine; INFORMATION sharing
- Publication
Traitement Automatique des Langues, 2020, Vol 61, Issue 2, p41
- ISSN
1248-9433
- Publication type
Article