Question Answering Integration

Using Buzzeasy you can integrate your custom question answering databases from Microsoft to your Buzzeasy digital channels.

Provision Language Question answering

This section of the documentation provides a comprehensive walkthrough for the Azure Resources provisioning. You can read more about Language custom questions through the official Microsoft documentation found here

To begin provisioning Language Question Answering, follow these steps:

Warning

You require Azure Portal permissions to perform these steps. Additionally, consult with your team about costs as Language Question Answering has its own cost.

  • Navigate to https://portal.azure.com
  • In the search bar of Azure Portal, type in Marketplace. This will take you to Azure's Marketplace.
  • In the Marketplace search bar, type in Language Service, then click Create.
  • Select the Custom question answering, then click the Continue to create your resource button located at the bottom left corner of your screen.
  • Select your azure Subscription, then select an existing resource group or create a new one. We recommend giving it a good name.
  • Select the region that is closest to your customers. Keep in mind that different regions might have different pricing options.
  • Provide a name for your new language service. Follow your internal Azure naming conventions, and keep your cloud environment clean!
  • Select your price tier. We do not recommend using an F0 plan for live customers, keep this for your internal testing.
  • Select the region for the Azure Search service, this will be created alongside your Language service.
  • Select the pricing for the Azure Search service. This varies based on your usage and amount of projects, also known as knowledge bases. We do not recommend using Free F for live customers.
  • Read the Responsible Use documentation, then click the checkbox to agree to the terms.
  • Your administrator may configure tagging policies, click the Tags tab to add the proper tags according to your internal policies.
  • Once you're all set, click the Review+Create button. Validation will begin.
  • If you passed all of your internal policies, validation will be green and the deployment will begin. This can take up to a few minutes.
Note

If you require assistance, Geomant engineers are happy to help you.

Once deployment is complete, you will need to grab 2 details required by Buzzeasy. In the Azure Portal search bar, type in the name of the Language Service created earlier. On the left side of the screen, click the Keys and Endpoint menu item, then click Show Keys. Copy Key 1, this will be your Buzzeasy Secret. Copy the Endpoint, this will be your Buzzeasy Endpoint URL.

Managing projects & knowledge bases in Language Question Answering

In Language Question Answering projects are a set of knowledge bases. In Buzzeasy you reference a project's name instead of a Knowledge Base.

To create a project for the first time, navigate to https://language.cognitive.azure.com, then log in with your Azure credentials.

Click the Create New button and select the Custom Question Answering option.

Select one of the two options presented, then provide your project's name, description, source language, and default answer. You may change your Azure Language Resource or the Azure search Resource as well.

Click Review, then click Create Project. Your project is now created! Copy the exact name, you will need this in Buzzeasy.

Now that you have a project, click on it, then click Add Source. You can fetch data from URLs, and files as well as built-in chit-chat knowledge bases.

You can add as many sources as your pricing allows you to.

Click on a source to begin reviewing, modifying, and testing questions. Once you are satisfied with the results, click Save.

On the left side of the screen, click Deploy Knowledge base. You must perform this step to push your changes to the live after testing them.

Adding Language Question Answering to Buzzeasy

Navigate to Buzzeasy's new portal and locate the Integration section. Click the QnA bots.

Click the Create button and fill in the details as follows:

  • Project Name: The name of your Language Question answering project (case sensitive).
  • Endpoint URL: Endpoint of your Language Service. Obtain this by searching for your Language Service in the Azure Portal > Keys and Secrets.
  • Secret: Key 1 of your Language Service. Obtain this by searching for your Language Service in the Azure Portal > Keys and Secrets.
  • Confidence threshold%: Percentage matching the knowledge base question. If the QnA bot falls below this threshold, it will exit the conversation. Read more about the confidence score through Microsoft's documentation. We recommend setting a 50-70% confidence threshold.

Using QnA in Buzzeasy Workflows

Buzzeasy Chat Workflows feature a QnA node type, this will allow you to select one of your existing QnA configurations and specify the next step in case the QnA falls below the confidence threshold and exits. Read more about Buzzeasy's chat workflows and the QnA node here.

Migrate from qnamaker to question answering

Microsoft provides extensive documentation on how to migrate from qnamaker.ai to Language Question answering.

To make things easier for you, this article provides a summary of how to migrate from qnamaker to Language Question Answering.

Follow these steps:

  • Assuming you already have a language service and search created as detailed earlier in this article, navigate to qnamaker.ai.
  • Click the Start Migration button.
  • Select your tenant.
  • Select the QnA maker resource you want to migrate.
  • Select the target Language resource.
  • Select the knowledge bases you want to migrate.
  • Click Next.

This might take up to a few minutes depending on the size of your knowledge base. Your knowledge base will be copied over to https://language.cognitive.azure.com/. We recommend reviewing the questions and answers pairs after the migration is complete.

For further instructions, please follow the official Microsoft documentation.

Agent Assistance

Buzzeasy allows administrators to configure a Question Answering to assist agents. Generally, Question answering is used to save agent time by providing answers to end-customers. However, Agents can benefit from the same functionality as well while engaged in a chat conversation with an agent.

You can specify a QnA configuration for a given queue node by setting the Agent Assistance field to the QnA configuration you wish to use. This configuration will only be used to assist agents. Should you want to use the same configuration to aid customers as well, simply add a QnA bot node to your workflow.

Limitations

The Buzzeasy Language Question Answering integration is supported for Webchat, Viber bots, and Facebook Messenger.