Predicting new topics

[VenkataBuddha]

Hi Everyone,
So when we use the model.transform(newdocs), we get the topics that were already generated/identifed in the pre trained model.
But what if we have a new topic in the newdocs. What can be done to assign a new topic to these new docs instead of them being assigned '-1'.

[MaartenGr]

Just to be sure I understand your question correctly, do you mean with predicting new topics that they fall in an outlier category, that they fall in a previously found topic, or that they may end up in an entirely new topic?

If it's about going from an outlier (-1) to a non-outlier topic, then your can use the function .reduce_outliers for that.

If it's about discovering entirely new topics, then it might be worthwhile to use online BERTopic for that which, combined with River, allows for discovering topics that were not found before.

[VenkataBuddha]

Thank You for the quick response Maarten!
Apologies for the confusion in my post, but yes I mean about docs ending up in entirely new topics.
I will reffer the online BERTopic documentation!