NLU bringing very wrong results with 100% reliability.

Albergaria
Tera Contributor

We have identified strange results using NLU. Some expressions that are not registered in the intent are achieved with 100% reliability. In addition, by using random letters we also hit some intentions the wrong way. Even giving negative feedback on these predictions, they keep showing up. We would like to know how to avoid these wrong results appearing in user searches.

 

I looked up "llllllllllll" and it found an intent with 100% reliability and no utterance of that intent has anything remotely like what I searched for:

VictorAlbergar_0-1680012828629.png

 

 

20 REPLIES 20

Hi Brian.

 

Thanks for posting the information. My model is doing the same thing. I looked at the changes you performed and does not seem to relate. If I enter a one word entry like "wordpress" or "notifications" it is always picking our password intent.

 

Any other suggestion of what would cause a single word to not be handled properly?

Thanks.

Brian Bakker
ServiceNow Employee
ServiceNow Employee

@k_lutz 

Single word utterances have no context and should be avoided when adding sample utterances in your intents. If the utterances "wordpress" and "notifications" are returning the Password intent, you will have single word utterances in this intent that are loosely related to these utterances. If these utterances should return other intents, add utterances containing "wordpress" and "notifications" to their respective intents that should be predicted, and remove any single word utterances in the Password intent, and add context to your sample utterances. You should also add a Vocabulary Item for "wordpress" with the synonym "Publishing platform". If these utterances should not return any intents, then use the Test panel feedback and select "No intent should be predicted" for the utterances "wordpress" and "notifications" that will add entries to table [ml_labeled_data] and ensures no intent is returned for these two utterances when using any NLU Model.

Thanks for the information Brian but I have tried those steps and it still goes to the password intent.

I have no single word utterances. I have flagged that it should have no intent. Please see screenshots. It does this for most single word entries or a single word will go to some other intent that has no relevance.

 

I have a case open with support but they just keep giving me links to KBs or DOC entries which I have already read and provide 0% help.

I am curious. Did you check the Expert Feedback Loop for Password Reset intent and see if "WordPress" is in there?

ServiceNow tells me that EFL is not used by the NLU but I have had experiences where user says something, VA gives a weird response, I go to EFL, find the phrase in there, tell it to ignore that phrase or move it to the correct intent and then the phrase picks up the correct intent.

This is a far fetched idea but worth a try.

Thanks Lynda1. I looked in there and did not see any match for wordpress or something similar. At this point, I don't really see the value in using NLU or the virtual agent. Based on feedback I have received, most users don't even try using the VA anymore.