I would like to be able to classify documents with multiple labels, and rank them, basically in situations where more than one label is appropriate. Is this possible?
Hi! The multi-label text classification workflows and interfaces currently assume that label applies or doesn't apply, which is also what most text classification models expect.
What type of model implementation are you looking to train on the ranked label information? And what structured data do you need to create for it?
And I guess a related question that probably has an impact on the type of workflow you choose for annotation: If there's an ambiguous case, would you rather have it annotated with label X and a very low rank, or not annotated with label X?
Great questions! I am doing intent classification. I was able to choose multiple labels for ambiguous cases, but I thought ranking them might be useful. Ranking would be for experimentation to address ambiguous cases and see if there are situations where two options would be relevant.