Zero-shot modeling unable to predict new documents.

[HaoyuWangEBSCO]

I am working on a topic modeling job for customer feedback, and I have found that when models are trained with zero-shot modeling, they become unable to predict or fit.transform new documents. Has anyone else experienced this, or does anyone know a way to avoid this?

[MaartenGr]

Have you tried installing BERTopic from its main branch? There are a number of fixes pertaining to zero-shot topic modeling.

[HaoyuWangEBSCO]

Hi Maarten, Thank you for your response! By that, do you mean to reinstall the BERTopic and re-training the zero-shot topic modeling model?

[MaartenGr]

You can do a pip install from the latest commit which should have the fix available. You can find more about that here. Then, you would indeed re-train the zero-shot topic model.