The aim of this project for the NIH Genetic and Rare Diseases Information Center (GARD) is to automatically identify rare disease articles as epidemiological, and if so, extract the epidemiological information from them. It is currently composed of two components:
- A long short-term memory recurrent neural network that classifies rare disease publications as epidemiological or not (created by Jennifer John, see more)
- A bidirectional transformer-based model that performs named entity recognition to identify epidemiological information from abstracts classified as epidemiological. (created by William Kariampuzha)
The EpiExtract4GARD folder contains the entirety of the transformer-based NER model. The api folder contains the API code and environments that runs the models.