Confidence score/ Alexa not responding right

Is there a way to adjust the threshold for the confidence score?

I am currently using an intent for the user to ask “what is my address”, it works fine and Alexa answers with users address stored in variables.


If I ask “what is my name” it also responds with the address, very DUMB. I am guessing because it is matching “what is my” with the address intent (since I don’t have intent for “what is my name”)

Is there a workaround to this?

There’s no something like confidence score.

I’m not sure about your skill inside, but “Address” is not suitable for slot, IMO.

  • There’s no suitable built-in slot types for whole address. There’s built-in slot types for AMAZON.Country, Amazon.City, AMAZON.Region, AMAZON.StreetName, although.
  • Generally “Address” has too many variations. It might works for you, but might not work for others. too many misunderstanding by alexa may frustlate users.
  • If you can get complete address from users, use Userinfo Block, then you can get users’ complete address without misunderstanding.

also, I guess your intent is like “what is my {slot}”. what slot type you used for this slot? if you don’t choose suitable slot, Alexa tries to match anyway as much as she can because there’s only a few hints for guess.

In my case, I’m not using the built-in slot feature for address. I am pulling the data in from spreadsheet integration, the user’s address is set as a variable when the user signs up (for example: when user rents a home and moves in, their new address is provided to me by the owner of the rental house). When a person moves in, they can use alexa, and a frequent question as a new mover is “what is the address here”

The issue I experience when testing is if I ask something similar using “what is” , “what is the” example “what is the nearest park” and if I don’t have an intent set for “nearest park” then it chooses my address intent as the closest match

my intent for address I did not use “what is my {slot}” as you mention. I simply used “what is my address” “what is the address here” “what is my new address” etc

I’m trying to see if there is a way to closely match the intent if “address” is found, and if not, it should go to “else”

I see. how many utterances you add? Basically Alexa do not try to fallback as much as she can and try to match one of the nearest intent based on utterances added in the intent.

If you add just a few utterances, alexa does not have enough hints to guess what user intents correctly.

In that case, you should add some more utterances.

Actually, I just tested in your similar situation.

  • only one intent for user to get an address, called “address_intent”
  • first, add utterances below
    • “where is my address”
    • “where is my new address”
    • “what is my address”
    • “what is my new address”
  • test: “where is the nearest park”
    • it goes “address_intent”
  • second, add some more intents.
    • “give me my address”
    • “give me my new address”
    • “I would like to know my new address”
    • “I want to know my new address”
    • “I would like to know my address”
    • “I want to know my address”
    • “address”
    • “new address”
    • “my address”
    • “my new address”
  • test again: “where is the nearest park”
    • it goes “else”

That means, with a few utterances, Alexa cannot recognize the differences between "“what is the nearest park” and “where is my address”. but adding more utterances, alexa will know its difference.

In my experiences (based on in Japan. I’m not sure if this is same in US or other countries, sorry), you have to add at least more than 20 utterances to pass Amazon’s certification.

so originally I did use 8-12 utterances per intent using long sentences, but then I ran into those issues because the similarity so I ended up reducing the amount of utterances as well as using keywords and not so much sentences. for example “my address” “address here” “shipping address” “new address” etc this way Alexa matches the keyword vs the similar words in a sentence

I also ended up creating a “negative word” intent basically using keywords that should not trigger any of my other intents e.g. “nearest park”. This seems to work better.

Curious to see now if adding more keyword phrases would help it like you said with using more

I did not know this! Do you mean at least 20 utterances per intent? or 20 utterances per skill?

sorry I didn’t explain it enough. This is a kind of better practice among developers in Japan who had some experiences for certifications, not Amazon’s official standard(Amazon has never say specific numbers). so I’m not sure if this is same as other locales. But one of typical reason of rejection was to be few utterances.

so, too many is better than too few.

This depends on situation, but I will introduce about my live skills.

  • average 15 utterances per custom intent
  • for maximum, a custom intent has 26 utterances
1 Like

You can also try to set the Skill Model Sensitivity to Medium or High in your Alexa project settings.