Skip to content

bot-tutorial-luis.md

Latest commit

 

History

History
143 lines (83 loc) · 5.24 KB

bot-tutorial-luis.md

File metadata and controls

143 lines (83 loc) · 5.24 KB
Raw

Using LUIS for Language Understanding

Up until this point, we have been using a simple regex recognizer to detect user intent. Bot Framework Composer has deep integration with LUIS.

Let's go ahead and update our dialog's recognizers to use luis instead.

Update recognizer

  1. Click on WeatherBot.Main in the left dialog navigation panel, go to the property editor for this dialog and locate the Recognizer type option and set it to LUIS

  2. To work with LUIS recognizer, you can provide content in the [.lu file foramt][2] that is highly similar to language generation format.

    With this, intents are denoted using the markdown section notation - e.g. # intentName and utterances are provided as a list.

    Replace the recognizer content with this -

    > See https://aka.ms/lu-file-format to learn about supported LU concepts.

> Weather intent and its utterances

# Weather
- get weather
- weather
- how is the weather

> Help intent and its utterances

# Help
- help
- i need help
- please help
- can you please help

> Cancel intent and its utterances

# Cancel
- cancel
- please cancel that
- stop that

  3. Once you have done this, you need to re-configure the various Handle an intent triggers within that dialog.

  4. Click on weather trigger in the left navigation and choose Weather from the intent drop down

    Update the title of the trigger to Weather instead of Handle an intent

  5. Click on cancel trigger in the left navigation and choose Cancel from the intent drop down

    Update the title of the trigger to Cancel instead of Handle an intent

  6. Given we are using LUIS which is a machine learning based intent classifier, we want to avoid low confidence results. To do this,

    Set the Condition property to this

    #Cancel.Score >= 0.8

    This says do not fire the cancel trigger if the confidence score returned by LUIS is lower than or equal to 0.8

  7. Click on help trigger in the left navigation and choose Help from the intent drop down

    Update the title of the trigger to Help instead of Handle an intent

    Set the Condition property to this

    #Help.Score >= 0.5

    This says do not fire the cancel trigger if the confidence score returned by LUIS is lower than or equal to 0.5

  8. Click on Restart bot

    Composer has now detected that you have LU information specified and it needs to create/ update corresponding LUIS applications.

    For Authoring key, click on environment in the top right corner of your screen (this is right above where this documenation content is displayed) and copy LUIS authoring key 1 into the composer window

  9. Click Publish. This should take a minute or two to complete. Composer will render progress at the top right corner of the screen.

  10. Click on Test in Emulator and talk to the bot.

With LUIS, you no longer have to type in exact regex patterns to trigger specific scenarios for your bot. Try things like:

  • "How is the weather"
  • "Weather please"
  • "Cancel everything"
  • "Not sure what I can do"

Using LUIS for entity extraction

You can use LUIS to recognize entities from what the user says.

As an example, the user could say "How is the weather in 98052?" and instead of prompting the user again for a zipcode, we could just go straight to pulling up weather information if the user had already provided that as part of their initial query.

Let's get this wired up.

  1. Step one is to add a regex entity extractor to the LUIS app. To do this, click on WeatherBot.Main and on the right side, add the following entity definition at the end of the LU content -

    > regex zipcode entity. Any time LUIS sees a five digit number, it will flag it as 'zipcode' entity. 

$ zipcode : /[0-9]{5}/

  2. Next, let's configure your prompt for zipcode to use the entity if LUIS finds it. To do this, open getWeather dialog from the left navigtation menu, select BeginDialog trigger and scroll to and find the prompt.

  3. Let's insert an action before the prompt by clicking on [+] -> Manage Properties -> Set a property

  4. Since the prompt itself is trying to set the zipcode in the user.zipcode, let's set that proeperty to the @zipcode entity.

    Set Property to

    user.zipcode

    Set Value to

    @zipcode

  5. Click on Restart bot, wait for the LUIS application to be updated (since you added a new entity) and then click on Test in Emulator

Now when you say "how is the weather in 98052" you should see your bot directly provide weather for that location instead of prompting you for zipcode.