Introducing Question Answering in Public Preview

by Contributed | May 31, 2021 | Technology

This article is contributed. See the original author and article here.

QnA Maker is an Azure Cognitive Service that allows you to create a conversational layer over your data in minutes. As part of our AI at Scale initiative across Microsoft, we are making the latest breakthroughs in natural language understanding available across our products and within our Azure Services portfolio. Powered by our Turing natural language model, we are excited to introduce the new Custom Question Answering feature (public preview) with Text Analytics. Alongside Custom question answering, we are also introducing a new Prebuilt Question answering capability that lets users extract relevant answers to questions from a given passage of text. 

Overview 

• Custom question answering supports all capabilities introduced with QnA Maker managed. 


• You can add unstructured files to your knowledge bases with Custom question answering. 


• We are also introducing Prebuilt question answering capability that lets users extract relevant answers to questions from a given passage of text. 

Enabling Custom question answering feature 

As Custom question answering is now a feature in Text Analytics, users should first visit the Text Analytics resource create blade on Azure portal. They will have the option to enable Custom question answering feature within the resource. Once the Custom question answering feature is selected, the user will update all details related to the feature in the create blade. On creation, a new Text Analytics resource with Custom question answering feature is deployed. 

If users don’t enable the feature on resource creation, they will have the option to enable the feature later through the Features tab of the Text Analytics resource blade. They will also be able to disable the Custom question answering feature via the Features tab. 

Support for unstructured documents 

With Custom question answering (preview), we are introducing the ability to add unstructured files to the knowledgebases. Now the content author can ingest entire documents in the knowledgebase and when a user query is passed, a response is returned on searching the documents ingested. The content authors don’t need to manage QnA pairs for unstructured documents.  

Take for instance the following unstructured document introducing Surface Products: SurfaceBlog. The user can ingest this document in the knowledge base and test for user queries from the text. We tested this with a few queries as shown below. 

Prebuilt Question Answering 

Prebuilt question answering provides users the capability to answer questions over a passage of text without having to create knowledge bases and manage additional storage. This functionality is provided as an API and can be used without having to learn the details about QnA Maker. Given a user query and a block of text/passage the API will return an answer and precise answer (if available). We currently support the ability to pass text from 5 documents. 

API Updates 

We have added a new preview release for v5.0 APIs: V5.0-preview.2. Users should access this version for unstructured documents support and pre–built question answering.  

Pricing 

Custom question answering feature is free in public preview. However, users will be charged for Azure cognitive search they add to the feature as per the tier selected. 

Note to existing QnA Maker managed users 

1. QnA Maker managed has been re-introduced as Custom question answering feature in Text Analytics. 


2. Going forward users will not be able to create a new QnA Maker managed resource. All new resources can be created by enabling Custom question answering feature with Text Analytics. 


3. All existing QnA Maker managed (preview) resources continue to work as before. The resource settings, portal, endpoints, keys, SDK, etc. pertaining to existing QnA Maker managed resources will work as is. There is no action required from the user at this point in time.  


4. We will continue to support v5.0-preview.1 for existing QnA Maker managed customers. 


5. If the users choose to migrate to Custom Question answering, they can do so. To migrate from QnA Maker managed to Custom question answering, users can create a Text Analytics resource with Custom question answering feature enabled and migrate the knowledge bases from QnA Maker managed to the Text Analytics resource. 


6. Custom question answering (preview) continues to be offered in free public preview. 


7. Custom question answering (preview) is available in the following regions: 
   
   
   
   • South Central US 
   
   
   • North Europe 
   
   
   • Australia East
   
   
   
   

References 

• Set up a QnA Maker service – QnA Maker – Azure Cognitive Services | Microsoft Docs 


• Configure QnA Maker service – QnA Maker – Azure Cognitive Services | Microsoft Docs 


• Manage knowledge bases – QnA Maker – Azure Cognitive Services | Microsoft Docs 


• Limits and boundaries – QnA Maker – Azure Cognitive Services | Microsoft Docs 


• Prebuilt API 

DISABLE_PARAMETER_SNIFFING

by Contributed | May 31, 2021 | Technology

This article is contributed. See the original author and article here.

According to MSDN article, ‘DISABLE_PARAMETER_SNIFFING’ instructs Query Optimizer to use average data distribution while compiling a query with one or more parameters. This instruction makes the query plan independent on the parameter value that was first used when the query was compiled. This hint name is equivalent to trace flag 4136 or Database Scoped Configuration setting PARAMETER_SNIFFING = OFF.

Looks like it’s a pretty good hint, However, it doesn’t means you can resolve all parameter sniffing issue by using this query hint.

Actually,  the sentence ‘Query Optimizer to use average data distribution while compiling a query with one or more parameters’ is not 100% correct. It really depends on what symbol you used in the where clause.

‘DISABLE_PARAMETER_SNIFFING’ is a replacement of variable, these two have exactly same effect. If you are not familiar with selectivity of variable, please review  my post Selectivity and Estimated Row: Variable – Microsoft Tech Community

I’m going to use AdventureWorks 2019 in this post.

——————–Please run this script—————

use AdventureWorks2019

go

IF exists(select 1 from sys.tables where name=’SalesOrderDetail’ and schema_id=schema_id(‘dbo’))

      drop table SalesOrderDetail

go

select * into SalesOrderDetail from [Sales].[SalesOrderDetail]

go

create statistics iProductID ON SalesOrderDetail(productid) with fullscan

Go

dbcc traceon(3604,2363)

——————–Please run this script—————

For example, following two stored procedure returns exactly same Estimated rows 456.

create proc ptest1

@pid int

as

select * from SalesOrderDetail where productid>=@pid option(use hint(‘DISABLE_PARAMETER_SNIFFING’))

go

create proc ptest2

@pid int

as

declare @pid1 int =@pid

select * from SalesOrderDetail where productid=@pid1

Go

Trace flag 2363 displays more detail about the selectivity.

———————————-trace flag 2363 output———————————-

Begin selectivity computation

Input tree:

  LogOp_Select

      CStCollBaseTable(ID=1, CARD=121317 TBL: Sales.SalesOrderDetail)

      ScaOp_Comp x_cmpEq

          ScaOp_Identifier QCOL: [AdventureWorks2019].[Sales].[SalesOrderDetail].ProductID

          ScaOp_Identifier COL: @pid

Plan for computation:

  CSelCalcHistogramComparison(POINT PREDICATE)

Loaded histogram for column QCOL: [AdventureWorks2019].[Sales].[SalesOrderDetail].ProductID from stats with id 3

Selectivity: 0.0037594

Stats collection generated:

  CStCollFilter(ID=2, CARD=456.079)

      CStCollBaseTable(ID=1, CARD=121317 TBL: Sales.SalesOrderDetail)

End selectivity computation

———————————-trace flag 2363 output———————————-

 121317*0.0037594=456

Please review Selectivity and Estimated Row: Variable – Microsoft Tech Community for other inequations.