How to use LUIS in Microsoft Bot Framework

In one of the previous articles, I have talked about how to start building a bot with Microsoft Bot Framework (MBF). It was a simple bot that asks a companion to do simple math exercises. The bot application works great but the process of conversation looks not very natural as the app understands only two phrases: the standard utterance for starting the conversation and answering the question with a number.
slack-bot
Also in the LUIS article, I have spoken about the human-computer interaction and one of the main problem of this process – understanding human language by computer. The solution for this problem was a service called LUIS. Let’s take a look at how we can combine both services to create a bot that will understand human language better.

Integration LUIS in MBF application
As a basis for our new bot, we will use MBF and LUIS applications from the previous articles and create a new dialog which will understand commands for managing temperature of the air conditioner. The good news is that MBF has built-in support for LUIS application which means that you don’t have to handle HTTP requests manually.

Steps:
1) To start working with LUIS we need to inherit our dialog from LuisDialog<object> and add LuisModel attribute.

[LuisModel("{model_id}", "subscription_key")]
[Serializable]
public class ConditionerDialog : LuisDialog<object>
{
 
}

2) To get model_id and subscription_key we need to login to LUIS portal and open our app. Then click publish and copy values from URL.
luis-endpoint
3) Then we can add two methods to our dialog and mark them with LuisIntent attribute. In our case LUIS will call ChangeTemperature method when phrase received from a user will be the part of “Change temperature” intent. The method that is marked with empty LuisIntent attribute will be a handler from all non-recognized utterances.

[LuisModel("81910f1b-af4a-4f13-9029-88f8a4c1e561", "48e5df8c9d3c41a39faa7cef422a4241")]
[Serializable]
public class ConditionerDialog : LuisDialog<object>
{
    [LuisIntent("Change temperature")]
    public async Task ChangeTemperature(IDialogContext context, LuisResult result)
    {
    }
 
    [LuisIntent("")]
    public async Task None(IDialogContext context, LuisResult result)
    {
    }
}

4) The result from LUIS with entities, intents and query is available in the result parameter. We write some code to add some logic to our dialog.

[LuisModel("81910f1b-af4a-4f13-9029-88f8a4c1e561", "48e5df8c9d3c41a39faa7cef422a4241")]
[Serializable]
public class ConditionerDialog : LuisDialog<object>
{
    private ConditionerState _conditionerState;
 
    [LuisIntent("Change temperature")]
    public async Task ChangeTemperature(IDialogContext context, LuisResult result)
    {
        if (_conditionerState == null)
            _conditionerState = new ConditionerState();
 
        var temperatureEntity = result.Entities.SingleOrDefault(e => e.Type == "builtin.number");
        double temperature;
 
        if (temperatureEntity == null || !double.TryParse(temperatureEntity.Entity, out temperature))
            await context.PostAsync("I can't find the temperature. Please enter a valid value.");
        else
        {
            _conditionerState.Temperature = temperature;
            var temperatureScaleEntity = result.Entities.SingleOrDefault(e => e.Type == "Temperature scale");
            var temperatureScale = _conditionerState.TemperatureScale;
 
            if (temperatureScaleEntity != null && !Enum.TryParse(temperatureScaleEntity.Entity, true, out temperatureScale))
                await context.PostAsync($"I didn't understand the {temperatureScaleEntity.Entity} temperature scale. Try to use Celsius or Fahrenheit.");
            else
            {
                _conditionerState.TemperatureScale = temperatureScale;
                await context.PostAsync($"Please wait a few minutes while AC will reach {_conditionerState.Temperature} {_conditionerState.TemperatureScale}");
            }
        }
 
        context.Wait(MessageReceived);
    }
 
    [LuisIntent("")]
    public async Task None(IDialogContext context, LuisResult result)
    {
        await context.PostAsync("Sorry I didn't understand that");
        context.Wait(MessageReceived);
    }
 
    [Serializable]
    public class ConditionerState
    {
        public double Temperature { get; set; }
        public TemperatureScale TemperatureScale { get; set; }
    }
 
    public enum TemperatureScale
    {
        Celsius,
        Fahrenheit
    }
}

5) The dialog is ready so let’s see how it works in Bot Framework Emulator.
bot-framework-emulator
You can see the utterances in LUIS portal. Also, you can review and fix LUIS suggestions. The more utterances you submit the more accurately LUIS recognizes your phrases.
luis-suggestions

Summary
With the help of Microsoft Bot Framework, you can create powerful bots that can be integrated into a lot of services like Skype, Facebook, Slack, etc. You can use LUIS service to help your bots understand a user better and become smarter with a time. Also, it’s great that MBF supports LUIS from the box and you just need to use it without care of manually handling HTTP requests.

How to use LUIS in Microsoft Bot Framework

Posted by Andriy Koval

Comments