Skip to content

Azure AI Language - Question Answering Solutions

🧩 Problem

You need to build an application that can automatically respond to user questions in natural language, such as an intelligent FAQ system, providing accurate and relevant answers.

💡 Solution with Azure

Use the question answering capability of the Azure AI Language service to create, test, and publish a knowledge base of question-answer pairs. This knowledge base can be integrated into applications or conversational bots.

🧱 Required Components

  • Azure AI Language resource
  • Azure AI Search resource (to index the KB)
  • Language Studio (for GUI-based creation and management)
  • REST API or SDK (optional, for programmatic access)

Data Sources:

  • Web pages with FAQs
  • Structured documents (e.g., brochures, user guides)
  • Built-in chit-chat datasets
  • (Optional) JSON file to define synonyms

🏗️ Architecture & Development

1. Create the Knowledge Base

  1. Sign in to Azure Portal
  2. Create a Language resource and enable question answering
  3. Associate it with an Azure AI Search index
  4. In Language Studio:
  5. Create a Custom question answering project
  6. Add data sources (URLs, files, chit-chat sets)

2. Implement Multi-Turn Conversations

  • Add follow-up prompts to questions to clarify user intent
  • Link follow-ups to existing answers or create new ones
  • Ensure responses are context-aware within multi-turn flow

3. Test and Publish the KB

  • Test questions interactively in Language Studio
  • Review confidence scores and possible answer alternatives
  • When ready, deploy the KB to a REST endpoint for external use

4. Use the Knowledge Base via API

Send HTTP POST requests with JSON like:

{
  "question": "What do I need to do to cancel a reservation?",
  "top": 2,
  "scoreThreshold": 20,
  "strictFilters": [
    {
      "name": "category",
      "value": "api"
    }
  ]
}

Response includes: matched question, answer, score, and metadata

5. Improve Performance with Active Learning

  • Enable automatic suggestion of alternate phrasings
  • Use Review suggestions pane to approve/reject them
  • Add alternate questions manually if needed

6. Enhance with Synonyms

Improve understanding of user input by submitting synonyms like:

{
  "synonyms": [
    {
      "alterations": ["reservation", "booking"]
    }
  ]
}

✅ Best Practices & Considerations

  • Combine question answering with language understanding when user input triggers different actions or intents
  • Use multi-turn interactions to clarify ambiguous queries
  • Regularly review suggestions from active learning
  • Add synonyms for domain-specific terminology
  • Define appropriate score thresholds for reliable responses

❓ Sample Exam Questions

Q: What is the primary difference between question answering and language understanding in Azure AI Language? → Question answering returns a static answer; language understanding identifies intents and entities.

Q: Which component is required to host the index for a custom question answering project? → Azure AI Search

Q: How can you improve the accuracy of a knowledge base over time? → Enable active learning and define alternate phrasings

Q: What does the scoreThreshold parameter control in a question answering API request? → Minimum confidence score required for an answer to be returned

Q: What is the purpose of defining synonyms in a question answering project? → To handle variations in terminology and improve answer matching