Code for Extracting-Social-Determinants-of-Health-N2C2 The prediction task A, B, and C involves identifying trigger and argument spans, normalizing arguments, and predicting links between trigger and argument spans.
- wheel
- pandas
- tqdm
- spacy>=3.0.0 with languange model "en_core_web_sm"
-
MIMIC-III
-
University of Washington (UW)
-
Training and test data for this task will utilize the SHAC annotations, which will be provided using the BRAT standoff format
-
Prepare data processed from BRAT
Run ./training_prepare.sh
and ./test_prepare.sh
We get the data under a directory with such setup:
# Description:
Triger: <Trigger><Drug>
Argument: <Argument><Drug><Type>
Annotation
└── triggers_tag
├── train
│ └── ann + txt -> conll -> combined
├── dev
└── test
└── arguments_tag
├── train
│ └── ann + txt -> conll -> combined
├── dev
└── test
Run ./model_train.sh
Run ./pipeline.sh