Cognitive Services QnA Maker – Create knowledge base from online FAQ

I want to describe how to create chatbot client communicate with the QnA Maker API. Although, at this article have to describe preparation for creating chatbot client app. If already created Microsoft Azure Cognitive Services QnA Maker, please skip this article then check next.
A chatbot responds using knowledge base when chatbot is asked some question. QnA Maker makes creating first knowledge base easily.

・Solution idea

Microsoft provides great guidance to design a solution architecture for chatbot which using QnA Maker API at this web site.

・Over view

A knowledge base creating feature of QnA Maker lead creating process as step by step.

・The analyze feature for FAQ web page

I wanna show you a sample to describe that the analyze feature for FAQ web page provided by the QnA Maker is so smart.

Rolling Stone, LLC, a subsidiary of Penske Business Media, LLC publishes article ‘Q&A: Eddie Van Halen‘ on the web site, I want to use this web site as sample to describe how useful the QnA Maker API is. This page is not create aim to work well online FAQ page analyze feature, because I chosen it as random one from search engine results. Nevertheless, the feature made very great results.

Below is a part of page source of the sample FAQ page.

<p>
<strong>Will old fans like this record?<br /></strong>
A lot of people hate it. It’s the third incarnation of the band. It’s not a departure,
but we’re branching out. I was listening to the radio yesterday and the DJ was just rambling over it,
saying Gary sounds like a weak Sammy Hagar. Then they took phone calls and one guy was like,
“Ah, I still like Dave.” So obviously he’s not gonna like it. What’s
funny to me is that this isn’t international espionage  — we’re
a fucking rock & roll band. People, get a life, will ya? If you can’t,
get on the Internet and join the club.
</p>

There is no ID attribute or Name attribute in the sample FAQ page source, but has only style class definition.

The online FAQ page analyze feature of the QnA Maker can understand that the content is question text using “?”. And also can understand that another part is the answer of it, with structure of the contents.

・Step 1 Create Microsoft Azure QnA Service of Cognitive Service

I created QnA Maker as free, F0 cost level and F search lebel. Sorry for Japanese screen shot.

・Step 2 Select items of ‘Step 1’

・Step 3 and 4 Define name and source

I defined used source as the name, this way is useful for maintenance such as relearning because correct answer ratio of chatbot is depended on knowledge base.

I used default text to Default answer text of multi-turn extraction. If the chatbot is used in specific circumstanse, change this text such as ‘do you want to know one of below ?’ because multi-turn extraction is used when chatbot selected multiple answers and needs to do continuous conversation.

And set online FAQ page address to URL.

・Chit-chat and Create

Using Chit-chat section to control atmosphere of conversation. This default set of Q and A helps establishing character of chatbot.(see ‘Add Chit-chat to a knowledge base‘ of Microsoft Docs)

・Test

About takao

I'm Microsoft MVP since June 2010.