The problem I'm working on involves span categorisation with spaCy, however some of the tags are ambiguous, e.g. span1 => 60% tag1, 40% tag2
I'm trying to figure out if there is a way to get Spacy's output to capture this ambiguity.
I have been looking at the model's confidence levels and at the moment Spacy's outputs are pretty much binary, e.g. span1 => 99% tag1, 1% tag2, which I guess makes sense in terms of its loss function.
Is there a way to modify the loss function so e.g. it only looks at recall rather than precision, or is there any other way I could approach this?
Thanks!