How to improve a supervised model by giving feedback to the model during prediction

[prabhatM]

The FastText does a wonderful job predicting the labels. But it's not always right. Is there anyway to tell the model if the prediction is right or wrong and if wrong what is the right label for that prediction through a CLI interface?

I might be suggesting adding a layer of "reinforcement learning" to improve the result.

It will help the technical domain specific data classification.

In highly technical domain a text "xxxxx WITH xxxxxxxx" and "xxxxx WITHOUT xxxxxxxx" have absolutely different meaning and we would like to retrieve the correct label for a query.

Some of the chat bot use this method to improve their model.

Or, am I missing something that FASTTEXT already does something of this kind?

[Celebio]

Hi @prabhatM ,
Currently it's not possible to update a trained model. You have to train from scratch with all your data combined.

Best regards,
Onur

[alext234]

It'll be nice if there is such option for incremental training though I'm not sure if it is feasible.