Is it possible to show the labeled entities from a NER model while annotating text segments for classification? Being able to see the entities from a model would greatly increase the speed of my annotations for classification. Maybe this is a very niche use-case, but if it's not currently possible, I would love to see this as a feature if it makes sense beyond the scope of what I'm doing.
Yes, if there are
"spans" present in the data you're streaming in, the
classification interfaces will highlight those in the text. So you could implement a slightly custom version of the
textcat.manual recipe (or any other workflow you're using) that processes the incoming texts with your NER model and adds the entities present in the
doc.ents to each outgoing example.
Here's the relevant code (which is also how recipes like
ner.correct do it):