This integration implements the Microsoft LUIS Language Understanding service. It utilizes the built-in NLP features part of Stealth 2.x. If you are still using Stealth 1.x, you will first need to upgrade to Stealth 2.x before you can use this integration.
For instructions on how to configure your Azure account signup for LUIS, please reference their docs. You won't have to set this anywhere, but this gem does utilize the latest v3 LUIS API version.
Once your account is setup, these are the configuration settings you will need to add to you services.yml file:
default: &default
nlp_integration: luis
luis:
endpoint: westus.api.cognitive.microsoft.com
app_id: 9434fbd8-420b-6d75-8a6f-b6c9a0ac5ec0
subscription_key: 1b69a4b9db669805b4fcba5f1f2f87bb
tz_offset: 0
intent_threshold: 0.2
production:
<<: *default
development:
<<: *default
test:
<<: *defaultStealth will automatically use your staging LUIS slot in development and staging environments and will use the production slot for your production Stealth environment.
That's it! Stealth will now automatically use LUIS for intent detection and entity extraction automatically via handle_message and get_match.
We recommend you name your intents using snake case (snake_case). This is because this integration will automatically convert your intent names to Ruby symbols.
So for example, if you have a handle_message defined like this:
handle_message(
'Maybe' => proc { step_to state: :say_maybe },
:yes => proc { step_to state: :say_yes },
:no => proc { step_to state: :say_no }
)If your user responds with a variation of the string maybe, then they will be taken to the state say_maybe.
Otherwise, the intent named yes and the intent named no will attempt to be matched. So if you had named your intent YES for example, you'd have to use :YES here which doesn't match Ruby syntax conventions.
For more info about how intents are matched, please see the Stealth NLP documentation.
This is the real number that respresents the minimum threshold required for an intent to match. So if for example your intent_threshold is set to 0.2, if the top intent scores 0.09, it will not be returned as a match.
The entity types listed below are named using their corresponding Stealth type. The equivalent type used by Microsoft LUIS is also listed. For each code sample, the sample query is first provided followed by the array of entities extracted from the queries (for the given type).
It's possible, and even likely, that a query matches more than one entity type. For example, a currency type will also match a number type. For more info about how to utilize these types, please see the Stealth NLP documentation.
LUIS prebuilt entity: number
"I think it was something like 63 or maybe 764"
[
63,
764
]"It was almost 15k"
[
15000
]For more info about these values, please reference the number entity LUIS documentation.
LUIS prebuilt entity: money
"send me $87 or 48 cents"
[
{ 'number' => 87, 'units' => 'Dollar' },
{ 'number' => 48, 'units': 'Cent' }
]For more info about these values, please reference the money entity LUIS documentation.
LUIS prebuilt entity: email
"you can contact me at john@email.none"
[
"john@email.none"
]For more info about these values, please reference the email entity LUIS documentation.
LUIS prebuilt entity: phonenumber
Note: LUIS does not parse nor attempts to clean up phone number.
"You can reach me at 313-555-1212"
[
"313-555-1212"
]For more info about these values, please reference the phonenumber entity LUIS documentation.
LUIS prebuilt entity: percentage
"The stock is up 8.9% today"
[
8.9
]For more info about these values, please reference the percentage entity LUIS documentation.
LUIS prebuilt entity: age
"81 years old"
[
{ 'number' => 81, 'units' => 'Year' }
]For more info about these values, please reference the age entity LUIS documentation.
LUIS prebuilt entity: url
"please visit google.com or https://google.com"
[
"google.com",
"https://google.com"
]For more info about these values, please reference the url entity LUIS documentation.
LUIS prebuilt entity: ordinalV2
"they finished 2nd and 5th"
[
{ 'offset' => 2, 'relativeTo' => 'start' },
{ 'offset' => 5, 'relativeTo' => 'start' }
]"she finished last"
[
{ 'offset' => 0, 'relativeTo' => 'end' }
]For more info about these values, please reference the ordinalV2 entity LUIS documentation.
LUIS prebuilt entity: geographyV2
"She moved to paris, france"
[
{ 'value' => 'paris', 'type' => 'city' },
{ 'value' => 'france', 'type' => 'countryRegion' }
]For more info about these values, please reference the geographyV2 entity LUIS documentation.
LUIS prebuilt entity: dimension
"it's about 4 inches wide"
[
{ "number": 4, "units": "Inch" }
]For more info about these values, please reference the dimension entity LUIS documentation.
LUIS prebuilt entity: temperature
"it feels like 98 degrees"
[
{ 'number' => 98, 'units' => 'Degree' }
]For more info about these values, please reference the temperature entity LUIS documentation.
LUIS prebuilt entity: datetimeV2
This one is the most complicated one to work with. The values are nested pretty deeply. This integration exposes the values at such a high level because there is a chance that LUIS will return results for more than one date type. For example, below we have just one result of type date, but LUIS could return more than one object of subtype daterange, time, timerange, etc. See the docs for more info about these subtypes.
"How about Mar 12?"
[
{
"type": "date",
"values": [
{
"timex": "XXXX-03-12",
"resolution": [
{
"value": "2019-03-12"
},
{
"value": "2020-03-12"
}
]
}
]
}
]For more info about these values, please reference the datetimeV2 entity LUIS documentation.
LUIS prebuilt domain entity: Calendar.Duration
"it will be between 15 minutes and 3 hours"
[
"15 minutes",
"3 hours"
]Additional docs for this prebuilt domain entitiy is not available
LUIS prebuilt entity: keyPhrase
"I need to find the instructional materials for the course"
[
"instructional materials",
"course"
]For more info about these values, please reference the keyPhrase entity LUIS documentation.
LUIS prebuilt entity: personName
"Little Cindy-Lou Who who was not more than two"
[
"Little Cindy-Lou"
]For more info about these values, please reference the personName entity LUIS documentation.