on 10-26-2020 11:59 AM
I've got an intent with 52 expressions, all based around the word "task manager".
The correct skill is triggered for basically any way to write this ("how do I...", "what is...", with and without typos), EXCEPT for the two-word expression "task manager" ("task manger" with typo works fine).
When the expression is "task manager", monitoring shows that the correct intent was recognized. But debugging shows that "multiple skills were triggered" (the correct one and two other ones, one of them: small-talk), and thus the fallback reply is used.
The triggers for the non-correct skills are simply the corresponding intents, and none of their expressions contains the words "task manager". I don't know why those skilly are triggered and I'm at loss how to fix this.
Thanks for your help!
Nina
Hello,
It should be a strictness issue. Even if the right intent is recognized, if the confidence is too low multiple skill as the fallback skill can be trigger. Be sure to have a high confidence on your intent detection and be sure that your skill are only trigger when this and only this intent is recognized.
If, after more training of your intent, the problem persist, you can add a confidence condition to your skill, to trigger them only when your intent is detected with a minimum confidence.
However, if you have created multiple skills, the good practice is to create at least the same amount of intents. If you add more intents to your dataset, confusion will be lower.
Best,
Doriane
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hello Doriane, thanks a lot for your reply!
I increased the strictness for the intended intent, but that didn't help:
In Monitoring, I see that "task manger" (with typo) is recognized with 67% confidence, triggering the correct response. And "task manager" (without typo) is recognized with 91% (!) confidence, but triggers the fallback due to "multiple skills triggered". In debugging, I cannot see the confidence regarding the 3 ("multiple") skills triggered.
Increasing the strictness more gradually prevents recognition of previously correctly "guessed" typo expressions, but doesn't improve performance for the non-typo expression. So I'm not sure if a confidence condition in the skill (or all 3 skills?) solves the problem.
Do I have to increase the strictness for the intents that feed the other two skills (and cause the fallback-due-to-multiple) too? Because quite many intents funnel into the smalltalk skill, so that would be a lot of work and might prevent "guessing" true smalltalk that's recognized with lower confidence.
I'm at loss here.
Thanks and best regards,
Nina
User | Count |
---|---|
75 | |
8 | |
8 | |
8 | |
7 | |
6 | |
5 | |
5 | |
5 | |
4 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.