Correct entity extraction but misclassified intents

Hello guys,

the entity extraction is working well but it is mapping to the wrong intent, to none. What could be the issue?


Serdar, It will be helpful if you can provide additional details like sample of your domain file, and sample story or flow to understand what exactly is happening here.

You might to run this in interactive mode first and see what’s going on. That may help you figure out the issue.

Thank you Apurva

thank you very much Apurva for getting to me pretty quick. I almost solved the issue, but I have just quick question. In each run of train, I am getting very different results for the test set. The performance metrics variation is very high for example the precision score of an intent is around 0.9 when i retrain the model it is getting to around 0.6. What could be the result? How can I get more robust results? Thank you very much Regards, Serdar

@serdar as far as I know some variation is expected (as with any ML re-training) but 0.9 to 0.6 seems bit too much. One possibility might be that test samples might be totally off from what model has learned. What does train/test split looks like? In my runs i do get pretty consistent results.

Thanks apurva again for getting to me back fast. Definitely I would expect some variation as well but as you mentioned not this much. The I am using a separate test file which is almost 10% of training set.

Regards Serdar