How to use Zero-Shot Classification with Open AI representation?

[imprateekagarwal]

I would like to assign a topic label from candidate topic if new document is similar to already generated topic labels otherwise would like to assign new labels using Open AI.

[MaartenGr]

You would need to update your prompt template such that it shows all candidate topics with the message to assign topics to documents in the list or create a new label. Do note though that this only works on topic-level since calling OpenAI for individual documents is quite inefficient.

[imprateekagarwal]

But can we handle it in chain of representation e.g.

representation_models = [PartOfSpeech("en_core_web_sm") , KeyBERTInspired() , mmr , zeroshot , openai_generator ]

[MaartenGr]

What the zeroshot gives to the openai generator are the assigned labels through [KEYWORDS], so you can make use of that. Personally, I would skip the zeroshot step and let the openai generator handle that. That would be much more efficient since the openai generator can also do zeroshot.