5.2.3. Question Answering and Text Analytics

Question Answering: Azure AI Language includes question answering capabilities for building knowledge base-powered Q&A systems.

How Question Answering works:

1. Create a knowledge base from FAQs, documents, or URLs
2. Train the service to understand your content
3. Query the knowledge base with natural language questions
4. Get ranked answers with confidence scores

Knowledge base sources:

• Existing FAQ documents or web pages
• Product manuals and documentation
• Manually created question-answer pairs

NOT supported as knowledge base sources:

• Audio files (must be transcribed first)
• Images (must be OCR'd first)
• Video content

Multi-turn conversations: Question answering supports follow-up prompts for multi-turn conversations:

• Define follow-up questions for ambiguous queries
• Guide users to more specific answers
• Handle complex queries across multiple exchanges

Text Analytics for Health: A specialized feature for medical text analysis:

• Extracts medical entities (medications, diagnoses, procedures)
• Identifies relationships between entities
• Links to medical ontologies
• HIPAA compliant processing

PII Detection Details: PII detection finds and can redact personally identifiable information:

Redaction options:

• Replace PII with entity type (e.g., "[PERSON]")
• Return character offsets for custom handling
• Different sensitivity levels available

Text Summarization: Azure AI Language can generate summaries of long documents:

• Extractive summarization: Selects key sentences from the original
• Abstractive summarization: Generates new summary text
• Useful for document processing, meeting notes, and content curation

⚠️ Exam Trap: PII detection can identify AND redact information. The service returns both the detected entities AND their positions in the text, enabling automated redaction workflows.