How can I explain and compare the results?

[premthomas]

Thank you for creating adding the function to use the Zero-Shot Classifier into transformer-interpret.

Could you please help me understand the scores of sentences (attribution score)?

From the documentation, it is clear that the scores of the tokens are normalized values (between -1 and 1) showing their contribution to the predicted class. In the visualization, we see that the sentence "Today apple released the new Macbook ... " is classified as "technology". Its attribution score is 1.43. The attribution score for "sports" is 1.61. Next to the labels, I also see the probability so I know which is the predicted class by looking at the table.

Is there any relationship between the word attribution scores and the sentence attribution scores? And how can I select the predicted class using the scores and not the probability values thereby being able to explain why one class was selected over another.

Thank you once again. And appreciate all the work you put into this.