Fig. 1: Disease-Specific Medical Ontology Learning (DiSMOL) Framework.

The disease-specific language model is trained on physician notes. Known symptoms, as informed by insomnia experts or existing instruments form the simple ontology, which is used as an input to the model to find contextually similar words. The resulting enriched ontology is then validated by an expert panel. Subsequently, the validated enriched ontology is utilized to transform physician notes into structured data, enabling the investigation of the disease burden.