Binary Classification Across Multiple Categories?

Good day everyone :slight_smile:

I am a complete novice, and not sure how to make an NLP model (or any other model for that matter) classify input data across Multiple Binary Categories, so I would appreciate some feedback (and not being demolished for asking dumb questions :hugs:)

Let’s say I have to classify sentiment of someone, from various interviews of them converted to text, along the categories of Ecstasy-Sorrow, Admiration-Loathing, and Humiliation-Pride, using the ULMFiT approach. There are multiple questions that arise from this:

  1. How do I set up multiple categories for binary classification? Can this same result be acheived with multi-label calssification?

  2. Can I just transcribe each interview and use them as inputs even though each would be well over 20k characers? Even if I can do that, is it the best way to train the model?

  3. Will using a key word to delineate speakers by placing that keyword at the beginning at their speech work in identifying to the model which speaker to classify?