Reducing outliers when outlier probabilities are all zero

[YanJ0201]

Hi,

I am trying to reduce outliers after training a BERTopic model. I am using reduce_outliers, but I noticed that the probability values for all my outlier documents are 0.

Because of this, I am not sure whether I should use the "probabilities" strategy. Is my understanding correct that if all outlier probabilities are 0, then probability-based reassignment cannot happen?

Does anyone have any insight into why the outlier probabilities might all be 0?

Thank you!

[MaartenGr]

Probability-based reassignment would indeed be difficult if they are all 0. Fortunately, there are many different strategies to use. Perhaps it might be worthwhile to recalculate the probabilities so you can use.