Troubleshooting for custom question answering

The curated list of the most frequently asked questions regarding custom question answering will help you adopt the feature faster and with better results.

Manage predictions

How can I improve the throughput performance for query predictions?

Answer: Throughput performance issues indicate you need to scale up your Azure AI Search. Consider adding a replica to your Azure AI Search to improve performance.

Learn more about pricing tiers.

Manage your project

Why is my URL(s)/file(s) not extracting question-answer pairs?

Answer: It's possible that custom question answering can't auto-extract some question-and-answer (QnA) content from valid FAQ URLs. In such cases, you can paste the QnA content in a .txt file and see if the tool can ingest it. Alternately, you can editorially add content to your project through the Language Studio portal.

How large a project can I create?

Answer: The size of the project depends on the SKU of Azure search you choose when creating the QnA Maker service. Read here for more details.

How do I share a project with others?

Answer: Sharing works at the level of the language resource, that is, all projects associated a language resource can be shared.

Can you share a project with a contributor that is not in the same Microsoft Entra tenant, to modify a project?

Answer: Sharing is based on Azure role-based access control (Azure Role-base access control). If you can share any resource in Azure with another user, you can also share custom question answering.

Can you assign read/write rights to 5 different users so each of them can access only 1 custom question answering project?

Answer: You can share an entire language resource, not individual projects.

The updates that I made to my project are not reflected in production. Why not?

Answer: Every edit operation, whether in a table update, test, or setting, needs to be saved before it can be deployed. Be sure to select Save after making changes and then re-deploy your project for those changes to be reflected in production.

Does the project support rich data or multimedia?

Answer:

Multimedia auto-extraction for files and URLs

• URLS - limited HTML-to-Markdown conversion capability.
• Files - not supported

Answer text in markdown

Once QnA pairs are in the project, you can edit an answer's markdown text to include links to media available from public URLs.

Does custom question answering support non-English languages?

Answer: See more details about supported languages.

If you have content from multiple languages, be sure to create a separate project for each language.

Manage service

I deleted my existing Search service. How can I fix this?

Answer: If you delete an Azure AI Search index, the operation is final and the index cannot be recovered.

I deleted my `testkbv2` index in my Search service. How can I fix this?

Answer: In case you deleted the testkbv2 index in your Search service, you can restore the data from the last published KB. Use the recovery tool RestoreTestKBIndex available on GitHub.

Can I use the same Azure AI Search resource for projects using multiple languages?

Answer: To use multiple language and multiple projects, the user has to create a project for each language and the first project created for the language resource has to select the option I want to select the language when I create a project in this resource. This will create a separate Azure search service per language.

Integrate with other services including Bots

Do I need to use Bot Framework in order to use custom question answering?

Answer: No, you do not need to use the Bot Framework with custom question answering. However, custom question answering is offered as one of several templates in Azure AI Bot Service. Bot Service enables rapid intelligent bot development through Microsoft Bot Framework, and it runs in a server-less environment.

How can I create a new bot with custom question answering?

Answer: Follow the instructions in this documentation to create your Bot with Azure AI Bot Service.

How do I use a different project with an existing Azure AI Bot Service?

Answer: You need to have the following information about your project:

• Project ID.
• Project's published endpoint custom subdomain name, known as host, found on Settings page after you publish.
• Project's published endpoint key - found on Settings page after you publish.

With this information, go to your bot's app service in the Azure portal. Under Settings -> Configuration -> Application settings, change those values.

The project's endpoint key is labeled QnAAuthkey in the ABS service.

Can two or more client applications share a project?

Answer: Yes, the project can be queried from any number of clients.

How do I embed custom question answering in my website?

Answer: Follow these steps to embed the custom question answering service as a web-chat control in your website:

1. Create your FAQ bot by following the instructions here.
2. Enable the web chat by following the steps here

Data storage

What data is stored and where is it stored?

Answer:

When you create your language resource for custom question answering, you selected an Azure region. Your projects and log files are stored in this region.