VA Topic Mapping to Intent even if It Should Not ..

rpriyadarshy · 3 weeks ago

Hi All

We have a Situation where - We expect a Given Topic should not be Shown in VA but Its Showing.

Example - User Wrote : Bitlocker Recovery ... VA showing Topic Named - Password Reset..

When Checked NLU Logs we see INTENT is there and also Entity with Value SOFTWARE with .99 Confidence Score...

SAMPLE LOGS

{
"utterance": "Bitlocker recovery",
"results": [

"scored_intents": [
{
"id": "Password Reset",
"name": "Password Reset",
"confidence": "0.8976258",
"prediction_quality": "HIGH"
}
],
"scored_entities": [
{
"id": "SOFTWARE",
"name": "SOFTWARE",
"value": "Bitlocker",
"normalization": "Bitlocker",
"confidence": "0.9999877",
"prediction_quality": "HIGH"
}

We do not expect this TOPIC to come Chat Bot. Since Playing with SYSTEM ENTITY may have some diverse impact. Please guide me with Some Solutions on this.

Regards

RP

Lynda1 · 3 weeks ago

This is going to sound strange, but this is what I discovered when our Virtual Agent gives the wrong topic to utterances.

I go to Expert Feedback Loop for VA

Select your model

Locate the intent that is getting picked up incorrectly - In your case the Password Reset

Look for any utterances related to Bitlocker

Select Mismatch - it will give you the option to select the proper intent.

Click Save feedback

View solution in original post

Lynda1 · 3 weeks ago

I have to work the Expert Feedback Loop for VA once a month since it runs every 30 days.

It does it's best matching phrases to intents and sometimes is does a poor job so I use the Mismatch and select Not relevant to any model.

Sometimes I can direct it to the correct intent if the intent is not close to the max.

I discovered, if I leave phrase in the EFL for VA when it is a correct match, the NLU uses that phrase.

When I see the VA pick up the wrong topics, the EFL is my first place to look at. Once I work it, the VA is corrected.