The project aims to classify a given text input of symptoms into a medical specialty that might be relevant to the symptoms. The model classifies the input into one of the 28 specialties, such as gastroenterologist, pediatrics, etc., that it was trained in.
This model is a fine-tuned version of
deberta-v3-small using the amazing
Blurr library (huggingface wrapper library using fastai)
The dataset for this task was built by scrapping
Patient.info. In the forum, posts are organized under different categories/groups such as anxiety disorder, acne, malaria, etc. These categories/groups were manually mapped to a specialty for eg anxiety to a psychiatrist, and malaria to an infectious disease specialist.
I have a blog on the same task using fastai's ULMFiT model. That can be found
here.
