import argparse
import spacy
import scispacy
from scispacy.linking import EntityLinker
from utils.get_text_from_csv import get_df, get_notes_single_row_id
def get_linked_entities(model, text):
print(f"Entity linking using {model}")
partial_input = '\n'.join(text.split('\n')[:10])
print(f"Input text (truncated): {partial_input}\n...")
output = {}
nlp = spacy.load(model)
# add the entity linking pipe to spacy pipeline
nlp.add_pipe("scispacy_linker", config={"resolve_abbreviations": True, "linker_name": "umls"})
doc = nlp(text)
# get named entities
ents = doc.ents
# each entity is linked to UMLS with a score
linker = nlp.get_pipe("scispacy_linker")
for entity in ents:
cur = {}
for umls_ent in entity._.kb_ents:
cur[umls_ent] = linker.kb.cui_to_entity[umls_ent[0]]
output[entity] = cur
return output
if __name__ == "__main__":
# parse command line arguments
parser = argparse.ArgumentParser(description='Linking named entities in MIMIC EVENTNOTES')
parser.add_argument('--mimic_dir', default='../../tutorials/data/mimic_data/', type=str, help='directory to mimic data')
parser.add_argument('--model', default='en_core_sci_sm', type=str, help='model for spacy')
parser.add_argument('--row_id', default=178, type=int, help='row id of text to be processed')
parser.add_argument('--output_file', default='./output_linked_entities.txt', type=str, help='output to save linked entities')
args = parser.parse_args()
mimic_dir = args.mimic_dir
model = args.model
row_id = args.row_id
output_file = args.output_file
print(f"Data file: {mimic_dir}NOTEEVENTS.csv")
df = get_df(mimic_dir + 'NOTEEVENTS.csv')
text = get_notes_single_row_id(df, row_id)
linked_entities = get_linked_entities(model, text)
with open(output_file, 'w') as f:
for entity, entity_dic in linked_entities.items():
f.write(str(entity) + '\n')
f.write('\n\n'.join([f"{k}\t{v}" for k, v in entity_dic.items()]))
f.write('\n\n\n')
print(f"Linked entities written to {args.output_file}")