My input utterance is scoring 100 but the Ranking and Resolver is discarding it stating verbMatchOnly. I am not sure why its getting discarded. Below is the snap from the debug log
“ml”: {
“definitive”: [
{
“task”: “get coverage status”,
“state”: “configured”,
“score”: 100,
“scoringCriteria”: “Fuzzy Search score”,
“matchType”: “definite”
}
],
“namedEntityRecognition”: []
}
R&R eliminates it with below reason
“eliminationInfo”: {
“reason”: “verbMatchOnly”
},
@aalekh.shrivastava
Let us know the below details -
- Utterance
- The task ML training
- ML Threshold settings
After responding here, please mention this post’s URL and send us the bot export if possible on koresupport@kore.com Or else if you are on our SaaS add koresupport@kore.com on the bot (add in admin as user first and then invite on the bot - Let us know your domain name and bot name).
Discarding a ML matched intent because of “verbOnlyMatch” means that the only thing from the ML training that the platform found matching the utterance was a verb, and verb only matches are not a great match.
You don’t mention what the utterance was or the training samples, but if the only utterance word that overlapped was, for example, “get”, then that is not a strong indicator - many intents could be include the word “get”.
The general solution would be to add more ML training to cover more words from the utterance, “get what”
I just wanted to add a few workarounds to this post in case someone sees the same behavior.
-
Disable rescoring of intents in R&R
-
Use patterns.
-
For single-word matching use a single pattern < ~someconcept > and just add the relevant words to that concept.
@andy.heydon Please feel free to share your thoughts if there can be more alternatives.
1 Like