Weak Supervision and Clustering-Based Sample Selection for Clinical Named Entity Recognition

Abstract One of the central tasks medical text analysis is to extract and structure meaningful information from plain-text clinical documents. Named Entity Recognition (NER) a sub-task extraction that involves identifying predefined entities unstructured free text. Notably, NER models require large amounts human-labeled data train, but human annotation costly laborious often requires training. Here, we aim overcome shortage manually annotated by introducing training scheme for uses an existing ontology assign weak labels provides enhanced domain-specific model adaptation with in-domain continual pretraining. Due limited resources, develop specific module collect more representative test dataset lake than random selection. To validate our framework, invite clinicians annotate set. In this way, construct two Finnish datasets based on records retrieved hospital’s evaluate effectiveness proposed methods. The code available at https://github.com/VRCMF/HAM-net.git .