ChatbotExample

Chatbot with Cognitive Services Personalizer

This sample shows how two Cognitive Services, LUIS and Personalizer, can be integrated into a coffee recommendation chat bot using ASP.NET Core 3.1.

Architecture overview

When a user starts up the chatbot, it sends them a welcome message with a randomly generated day of the week and the current weather status; within the chatbot implementation, these two features are used as context features for Personalizer, i.e. Personalizer will inform its decisions on what drinks to recommend the user based on what it has experienced in the past when under those same conditions. These features will be reset every time the bot is restarted or when the user prompts a reset. When the user replies back, the bot uses LUIS's intent recognition capabilities to discern what the user wants; the intents this LUIS app has been trained to recognize include showing the user a coffee and tea menu, asking the chatbot for a drink suggestion, approving or disproving of the suggestion, and resetting the context features. When a user expresses an intent that LUIS interprets as asking for a suggestion, the bot will gather the the context features mentioned above, as well as its list of drinks and their features, and send that information to Personalizer as a rank request. Personalizer will reply back with the same list of drinks and their corresponding ranking (i.e. how well Personalizer thinks the drink suits the user and the current scenario based on the features of the drink); the first item in the list is the top ranked suggestion, and the bot will in turn suggest it to the user. The user can then reply back. If LUIS registers their intent as liking the suggestion, the bot will send a reward request to Personalizer with the reward set to 1, indicating a positive response. Conversely, if LUIS interprets the user's reply as disliking the suggestion, the bot will send a reward request with the reward set to 0, indicating a negative response. Personalizer uses this feedback to update its model, which will affect how it responds in similar future scenarios.

Setting up the bot

Install Prereqs

• ASP.NET Core 3.1
• Microsoft Bot Framework Emulator is a desktop application that allows bot developers to test and debug their bots on localhost or running remotely through a tunnel.

Download the sample

• Clone the samples repository

git clone https://github.com/Azure-Samples/cognitive-services-personalizer-samples.git

• Open ChatbotSample.sln in Visual Studio or VS Code

LUIS Migration

• If you have a Language Understanding (LUIS) resource already created that you'd like to use, make sure you follow the LUIS migration steps here if any of the scenarios listed apply.

Necessary Azure resources

• Create both a Personalizer and a Language Understanding resource in the Azure portal. These will be used to power the reinforcement learning and NLP cababilities of the bot, respectively.
• NOTE: when creating a LUIS resource in the portal, you can choose between creating a prediction resource, an authoring resource, or both. You can opt to only use an authoring resource for this sample; however, the prediction key and endpoint that luis.ai will provide for you in this case can only be used for a limited number of requests. For a more robust app experience, you should create both an authoring and prediction resource. More on the two here.

Set up Personalizer

• In appsettings.json, modify PersonalizerServiceEndpoint to be your resource's endpoint and PersonalizerAPIKey to be either one of your resource's keys; this can be found in the Keys and Endpoint tab in the portal under Endpoint and Key1/Key2, respectively.

Set up LUIS

• Navigate to LUIS portal.

• Click the Sign in button.

• Click on My Apps.

• Select the Subscription for your LUIS authoring resource and then the LUIS authoring resource itself.

• Click on the New app for conversation button.

• Click on Import as JSON and select coffeebot.json from the cognitive-services-personalizer-samples/samples/ChatbotExample/CognitiveModels folder.

• Click on the Train button on the top right of the page; this will train your app on the model inside of chatbot.json.

• Once this is done, press the Publish button and select "Production" (You can leave the Sentiment Analysis, Spell Check, and Speech Priming fields set to Off). NOTE: Once you publish your app on LUIS portal for the first time, it takes some time for the endpoint to become available, about 5 minutes of wait should be sufficient.

• Update appsettings.json with your LuisAppId, LuisAPIKey, and LuisAPIServiceEndpoint, all of which can be found under the Manage tab for your resource:

  • LuisAppId can be found under Settings
  • If you created both a LUIS authoring and prediction resource:
    • Under Azure Resources > Prediction Resources, click Add prediction resource
    • Fill out the Subscription and LUIS Resource fields with your prediction resource, and click Done
  • LuisAPIKey can be found under Azure Resources > Prediction Resources listed as Primary Key
  • LuisServiceEndpoint can be found under the Azure Resources > Prediction Resources listed as Endpoint URL.

• You should now have an appsettings.json that looks like this:

  {
      "PersonalizerChatbot:" {
          "LuisAppId": "<your app ID>",
          "LuisAPIKey": "<your LUIS API key>",
          "LuisServiceEndpoint": "<your hostname, e.g. https://westus.api.cognitive.microsoft.com>",
          "PersonalizerServiceEndpoint": "<your Personalizer endpoint, e.g. https://myPersonalizerResource.cognitiveservices.azure.com/",
          "PersonalizerAPIKey": "<your Personalizer API key>"
      }
  },

Running and interacting with the bot

Visual Studio

• Press the F5 key to run the project

Visual Studio Code

• Bring up the terminal, navigate to cognitive-services-personalizer-samples/samples/ChatbotExample/ folder.
• Type dotnet run.

Connect to bot using Bot Framework Emulator

• Launch Bot Framework Emulator
• File -> Open Bot
• Enter <your App URL>/api/messages as the Bot URL. You can find your App URL by going to launchsettings.json where it is listed as applicationUrl. The default for this sample is http://localhost:3978

Interact with Chatbot

The chatbot can understand human inputs by defined "Intents" and "utterances". Those definitions can be found in coffeebot.json. For example, if you type "what do you suggest", LUIS will be able to categorize it as the Intent - "ChooseRank".

Intent Processing

In this example, LUIS Intents will be processed in PersonalizerChatbot.cs and then it will call the Ranking API to get a suggestion, shown below:

public async Task OnTurnAsync(ITurnContext turnContext, CancellationToken cancellationToken = default(CancellationToken))
{
    if (turnContext.Activity.Type == ActivityTypes.Message)
    {
        var recognizerResult = await await _coffeeRecognizer.RecognizeAsync(turnContext, cancellationToken);
        var topIntent = recognizerResult?.GetTopScoringIntent();
        if (topIntent != null && topIntent.HasValue && topIntent.Value.intent != "None")
        {
            Intents intent = (Intents)Enum.Parse(typeof(Intents), topIntent.Value.intent);
            switch (intent)
            {
                ...
                case Intents.ChooseRank:
                    var response = await ChooseRankAsync(turnContext, _rlFeaturesManager.GenerateEventId());
                    await turnContext.SendActivityAsync($"What about {response.RewardActionId}");
                    break;
                ...
            }
            ...
        }
        ...
    }
    ...
}

Deploy this bot to Azure (optional)

Follow this tutorial if you want to deploy your bot to Azure.

Further reading

• Azure Bot Service
• Personalizer Documentation
• LUIS Documentation