Fig. 2
From: Ontology-driven and weakly supervised rare disease identification from clinical notes
Weak supervision process for Text-to-UMLS linking. The left four white text boxes displayed the metadata (with examples) of a candidate mention-UMLS pair, identified by a Named Entity Recognition and Linking (NER+L) tool, SemEHR; the coloured text boxes in the middle show the contextual representation block (in
) and the rule-based weak data labelling (in
. A binary label is then generated, which weakly estimates whether the candidate pair indicates a correct phenotype of the patient. A phenotype confirmation model (in
) is then learned to select correct phenotypes from the pairs. (Figure adapted from [7]))