What is the most appropriate way of determining possible category/class/topic names for the zero shot classifier?
In other words, how does one go about finding what values should be entered in classes_verbalized list? Are potential labels completely left to human curation?
I have successfully used the following multlingual model to classify some non English texts: MoritzLaurer/bge-m3-zeroshot-v2.0 · Hugging Face
I used following ad-hoc category names:
classes_verbalized = ["politics", "economy", "entertainment", "environment", "health", "sports", "technology", "music", "art", "science", "history"]
What I am looking for is some sort of list or approach to generating labels.
How about binary classification (ie “hotdog” and “not hotdog”), what should my labels be then?
For example, I have a binary classification task to label “music review” documents. what should the other label be to represent “not music review”?