Embeddable AI

 View Only

FAQ Extraction via Watson Assistant Search Skill

By Tim Lai posted Wed December 02, 2020 02:42 PM


One of the common use cases for a Watson Assistant virtual agent (chatbot) is to provide responses to Frequently Asked Questions.  Currently, FAQs can be implemented in a variety of ways including:

  • Watson Assistant Intents for each common FAQ
  • A web crawl or document upload through Watson Discovery with additional configuration to specify page fields, Smart Document Understanding, etc

A new beta feature in Watson Assistant as of Q4 2020 called FAQ Extraction makes this a whole lot easier.  This FAQ Extraction feature allows for an easy configuration of Watson Discovery through the Search Skill to web crawl existing FAQ sites with minimal configuration.

To configure the FAQ Extraction feature, start from Watson Assistant and create a Search Skill (available in Watson Assistant Plus Plans and above)

Select your Watson Discovery instance.  If you don’t have a Discovery instance, provision one first before this step.  Click the “Create new collection” link

If you are testing Discovery with a Lite Plan, you may be prompted to upgrade to a more Production appropriate level plan.  To continue with this, you can either upgrade or click the “Let’s get started” link if you are using a Lite Plan.  Select the “Web Crawl” link on the next page:

Enter your URL for your FAQ page and click the Add button.  For this example, I am using this FAQ html example on a github repository.  Select how often you want the FAQ page to synchronize and the language. 

Click the “Apply FAQ Extraction” checkbox.  Note this button is only available if you have reached this page through the Watson Assistant Search Skill.  Click Save and Sync.  You will be sent to the Discovery page that says “processing your data”, this may take a few min or longer depending on the size of your FAQ page. 

Once the Discovery page is complete, it should show the number of documents processed which should match the number of FAQs you have. 

Click on the “Finish setup in Watson Assistant” button at the top right

Back in your Search Skill page, select your Discovery instance again and select the Collection name you just created.  Click the “Configure” button

The Search Skill page will read through your FAQ Collection and set the Question as the Title field, the Answer as the Body field, and a URL to the FAQ page.  This automatic detection of Question and Answer fields makes it quick to configure.

You can scroll down to the bottom for additional settings including how many Search Skill results to show (default is 3), as well as the new result selectivity setting which you can test to see how your results are affected.  Click the “Create” button to complete this FAQ Search Skill configuration

There are 2 additional steps required in your main Dialog Skill and the Assistant level

In your Dialog, either at the “Anything else” node or another node of your preference, select the “Search skill” under “Assistant responds” to call the Search Skill.  Typically, the “Anything else” node is where you will want to call a Search Skill if any of the previous Dialog nodes are not matched, the Search Skill serves as a fallback option for Long Tail searches.

At the Assistant level, create an Assistant and link your Dialog Skill and Search Skill together:

That’s it for the configuration of your FAQ Extraction.

To test it, you can then click the “Preview Link” tab and click on the preview URL.  Type in some text that will trigger the FAQ Search and you should see results such as this.

As you can see, the new FAQ Extraction feature makes it quick to configure your Watson chatbot to search against a predefined FAQ page without any coding or page field definition steps.