Simple Bot Creation with QnA Maker
Note: this blog assumes you have used Azure to create services in the past
One of the the most compelling scenarios for a bot is to add it to Facebook. A Facebook page is rather static. Finding information about a business on a Facebook page can be a bit of a challenge. And while users can comment, or send a message, the only replies they’ll ever receive is from a human, meaning the owner of the small business needs to monitor Facebook.
Of course, if it’s a small business that the page is representing, there’s a good chance the business doesn’t have the resources to create a bot on their own. Or, even if the business is of a size where they have access to developers, the developers aren’t the domain experts - that’s the salespeople, managers, or other coworkers.
To make a long story short, developers are often required to create the bot, and build the knowledge base the bot will be using to provide answers. This is not an ideal situation.
Enter QnA Maker.
QnA Maker is a service that can look at an existing, structured, FAQ document, and extract a set of question and answers into a knowledge base. The knowledge base is editable through an interface designed for information workers, and is exposed via an easy to call REST endpoint for developers.
To get started with QnA Maker, head on over to https://qnamaker.ai/. You can create a new service by clicking on New Service. From there, you’ll be able to give your service a name, and point to one or more FAQ pages on the web, or a document - Word, PDF, etc. - containing the questions and answers. After clicking create, the service will do its thing, creating a knowledge base that can be accessed via the endpoint.
The knowledge base is a set of questions and answers. After creating it, you can manage it much in the same way you edit a spreadsheet. You can add new pairs by clicking on Add new QnA pair. You can also edit existing pairs in the table directly. Finally, if you wish to add a new question to an existing answer, you can hover over the question on the left side, click the ellipsis, and choose Add alternate phrasing.
One important thing to note about the knowledge base, is each question and answer is an individual entity; there is no parent/child relationship between multiple questions and a single answer. As a result, if you need to provide additional ways to ask a particular question with the same answer, you will need to have multiple copies of the same answer.
Once you’re happy with the first version of your knowledge base, click Save and retrain to ensure it’s up to date. Then, click Test on the left bar, which will present you with a familiar bot interface. From this interface, you can start testing your bot by typing in questions and seeing the various answers.
You’re also able to update the knowledge base from this interface. For example, if you type a question that’s a little ambiguous, the interface will show you multiple answers on the left side. You can simply click the answer you like the most to update the knowledge base to use that answer for the question you provided.
In addition, after asking a question, and being provided an answer, you can add additional phrasings of the same question on the right side.
First and foremost, remember the eventual user experience for this knowledge base is via a bot. Bots should typically have personality, so don’t be afraid to modify some of the answers from their original form to make it read a bit more like a human typed it out, rather than a straight statement of facts. In addition, make sure you add multiple questions related to hello, hi, help, etc., to introduce your bot and help guide your user to understand the types of questions your knowledge base can answer. Finally, remember that while a single form of a question works well on a FAQ page, users can type the same question in multiple forms. It’s not a bad idea to ask other people to test your knowledge base to ensure you’re able to answer the same question in multiple forms.
And, once you’re ready to make the service available to a bot, click Save and retrain, and then Publish.
QnA Maker exposes your knowledge base as a simple REST endpoint. You can access it via POST, passing a JSON object with a single property of question. The reply will be a JSON object with two properties - answer, which contains the answer, and score, which is a 0-100 integer of how sure the service is it has the right answer. In fact, you can use this endpoint in non-bot services as well.
Of course, the goal of this blog post is to show how you can deploy this without writing code. To achieve that goal, we’re going to use Azure Bot Services, which is built on top of Azure Functions. Azure Bot Services contains a set of prebuilt templates, including one for QnA Maker.
In the Azure Portal, click New, and then search for Bot Service (preview). The Azure Portal will walk you through creating the website and resource group. After it’s created, and you open the service, you will be prompted to create a bot in Bot Framework. This requires both an ID and a key, which you’ll create by clicking on Create Microsoft App ID and Password.
IMPORTANT: Make sure you copy the password after it’s created; it’s not displayed again! When you click on Finish and go back to Bot Framework, the ID will be copied automatically, but the key will not.
Once you’ve entered the ID and key, you can choose the language (C# or NodeJS), and then the template. The template you’ll want is Question and Answer. When you click Create bot, you’ll be prompted to select your knowledge base (or create a new one).
And that’s it! Your bot is now on the Bot Framework, ready for testing, to be added to Skype, Facebook, etc. You now have a bot that can answer questions about your company, without having to write a single bit of code. In addition, you’ll be able to allow the domain experts update the knowledge base without any need for code updates - simply save and retrain, then publish, and your bot is updated.
While the focus has been on a no-code solution, you are absolutely free to incorporate a QnA Maker knowledge base into an existing bot, or to update the bot you just created to add your own custom code. And if you’re looking for somewhere to get started on creating bots, check out the Bots posts on this very blog, or the MVA I created with Ryan Volum.