A question about experimental results about triggerner HOT 3 CLOSED

Hi, @speedcell4
Thank you for your question.

We couldn't conduct experiments using more than 20% of sentences labeled with both entity tags and trigger tags due to annotation cost.
From our F1 curve, we could expect that using both entity tags and trigger tags will have better performance than just only using entity tags.

However, at the end like 100% sentences labeled with both entity tags and trigger tags,
we think the improvement might be very marginal since the NER model still affects a lot from entity coverage.
We argue that this model is more useful in a low-resource NER environment.

from triggerner.

sorry, mistakenly clicked the "Close issue" button

from triggerner.

Thank you for your response~

from triggerner.