Test out how well the ML model abstains from applying labels to abstracts that don't belong within the biomimicry taxonomy.

[bruffridge]

Summary of Issue

So we know the labeller is 53% accurate at labelling abstracts correctly from our training dataset which only contains abstracts that belong within the biomimicry taxonomy. I wonder how well it does at abstaining when given an abstract that does not?

[shrutix]

Starting with identifying a confidence/relevancy score for each abstract that is predicted with a label.

@pjuangph says: "would you be able to have your model export a confidence or let us see how it predicts with a test article? It could even be a copy paste string or read from a text file then send to the model for prediction. We want to see the output that it gives. "

[bruffridge]

related to #38