6.1. Analyze and Translate Text

💡 First Principle: Unstructured text hides structured information—customer reviews contain sentiment, contracts contain entity names, articles contain key topics. Text analytics extracts this structure, but different extractions answer different questions: "How do customers feel?" → sentiment. "What people/places/orgs are mentioned?" → NER. "What's this about?" → key phrases. The exam tests whether you can match the business question to the correct analysis type.

What breaks without proper feature selection: Sentiment analysis won't extract entities. Entity recognition won't give you sentiment scores. Key phrase extraction won't tell you whether the text is positive or negative. Each feature targets a specific extraction task, and using the wrong one gives you data you can't use—or worse, misleading insights that drive bad business decisions.

Consider a product review system: "The battery life is amazing but the camera is terrible." Simple sentiment might say "mixed." Opinion mining with show_opinion_mining=True tells you what is positive (battery) and what is negative (camera). That's actionable intelligence for product teams. The exam often presents scenarios where the answer hinges on choosing opinion mining vs. basic sentiment.

Service selection framework:

• Extract meaning from text → Azure AI Language (Text Analytics)
• Convert text between languages → Azure AI Translator
• Both in one pipeline → Translate first, then analyze (order matters for language-specific features)