5.1.1. Key Phrase Extraction

Key phrase extraction identifies the main topics and concepts in text. Imagine reading a 50-page report and highlighting just the 10 most important phrases—that's what this capability does automatically. It answers: "What is this text about?"

Key characteristics:

• Input: Text document or passage
• Output: List of important phrases (not sentences, just key terms)
• Used to summarize main topics quickly without reading entire documents

How it works: The algorithm identifies noun phrases and terms that appear significant based on frequency, position, and linguistic patterns. Unlike summarization (which produces sentences), key phrase extraction returns individual terms and short phrases.

Example output:

• Input: "Azure Machine Learning provides automated machine learning capabilities that help data scientists build models faster."
• Key phrases: "Azure Machine Learning", "automated machine learning", "data scientists", "models"

Common scenarios:

• Analyzing customer feedback for trending topics
• Summarizing meeting notes automatically
• Categorizing and tagging support tickets
• Building searchable document indexes
• Identifying themes across large document collections

What it does NOT do:

• Generate summaries (that's text summarization)
• Identify sentiment (that's sentiment analysis)
• Find named entities like people or places (that's entity recognition)

Key phrase quality factors:

• Relevance: Are the phrases central to the document's meaning?
• Completeness: Do they capture all major topics?
• Specificity: Are they specific enough to be useful?

Limitations to understand:

• May miss context-dependent importance
• Frequency isn't always the best indicator of importance
• Domain-specific terms may be over or under-weighted
• Very short documents may not have enough content for meaningful extraction

API response structure:

{
  "documents": [{
    "id": "1",
    "keyPhrases": ["Azure Machine Learning", "automated ML", "data scientists"]
  }]
}

Best practices:

• Use with other NLP features for comprehensive analysis
• Combine with entity recognition for complete information extraction
• Consider document length—longer documents yield better results
• Review extracted phrases for domain relevance

⚠️ Exam Tip: Key phrase extraction returns TOPICS and CONCEPTS, not named entities. If a question asks about extracting company names or people, that's entity recognition, not key phrase extraction.