The general idea is to classify objects that the model has never seen before, or in other words, allow us to assign an appropriate label to a piece of text without having received any training examples before. This approach has is also called zero-shot learning.
There are multiple advantages, including:
...