negspacy pairing example

[jenojp]

negspacy pairing example

[jenojp]

No problem! So I'm not sure how familiar you are with the inner workings of spacy but it has a really nicely designed way of processing text as a pipeline. If you're newer to spacy, I 100% recommend the course by @ines as it does a wonderful job of explaining this. So in this case, negspacy is a custom pipeline component that adds an extension to spans. (Entities in spacy are treated as spans).

What I meant in the README was that you can use the scispacy language model and their UMLS entitylinker in conjunction with negspacy to find specific UMLS concepts that are or are not negated.

For example:

import spacy
import scispacy
from negspacy.negation import Negex
from scispacy.umls_linking import UmlsEntityLinker

nlp = spacy.load("en_core_sci_sm")
linker = UmlsEntityLinker()
#this can take a while the first time
nlp.add_pipe(linker)

negex = Negex(nlp, chunk_prefix = ["no"])
nlp.add_pipe(negex, last=True)

Then if you process a doc, you could access two new custom extensions ._.negex and ._.umls_ents

doc = nlp("Patient denies cardiovascular disease but has headaches. No history of smoking. Alcoholism unlikely. Smoking not ruled out.")

for e in doc.ents:
    print(e.text, e._.negex, e._.umls_ents)

There are many different spacy pipeline components here

[kaushikacharya]

from scispacy.umls_linking import UmlsEntityLinker

As per the current release of scispacy, it should be

from scispacy.linking import EntityLinker

And

linker = UmlsEntityLinker()

should be

linker = EntityLinker(resolve_abbreviations=True, name="umls")

[aficionadoai]

Thank you very much @jenojp