Building a Text Analysis App

Building with text analysis

Two approaches to text analysis

Two ways to build text analysis apps

Feature	Azure AI Language (Foundry Tools)	GPT-4o (Chat Completions)
How it works	Dedicated NLP API with specific endpoints for each task	Send text to GPT-4o with instructions to analyse it
Output format	Structured JSON with scores and labels	Natural language response (flexible format)
Best for	High-volume processing, consistent structured output	Flexible analysis, combined with other tasks
Cost	Lower per-transaction for dedicated tasks	Higher per-token but more versatile
SDK	azure-ai-textanalytics	azure-ai-projects / openai

Approach 1: Azure AI Language SDK

Sentiment analysis

from azure.ai.textanalytics import TextAnalyticsClient
from azure.core.credentials import AzureKeyCredential

client = TextAnalyticsClient(
    endpoint="https://your-language-resource.cognitiveservices.azure.com/",
    credential=AzureKeyCredential("your-key")
)

documents = [
    "The product is amazing! Best purchase I've ever made.",
    "Delivery was late and the packaging was damaged.",
    "I received my order on Tuesday."
]

result = client.analyze_sentiment(documents)
for doc in result:
    print(f"Sentiment: {doc.sentiment}, Scores: pos={doc.confidence_scores.positive:.2f}, neg={doc.confidence_scores.negative:.2f}")

Key phrase extraction

result = client.extract_key_phrases(documents)
for doc in result:
    print(f"Key phrases: {', '.join(doc.key_phrases)}")

Entity recognition

result = client.recognize_entities(documents)
for doc in result:
    for entity in doc.entities:
        print(f"  {entity.text} ({entity.category})")

Approach 2: Using GPT-4o for text analysis

response = chat.complete(
    model="gpt4o-deployment",
    messages=[
        {"role": "system", "content": "Analyse the following text. Return JSON with: sentiment (positive/negative/neutral), key_phrases (list), entities (list with name and type)."},
        {"role": "user", "content": "Dr. Sarah Chen from MediSpark reviewed the order placed on April 15th for $4,500 worth of medical supplies. She was very pleased with the quality."}
    ],
    temperature=0
)

💡 When to use which approach

Use Azure AI Language when:

• Processing large batches of text (thousands of documents)
• You need consistent, structured JSON output
• Cost per transaction matters
• You need specific NLP features like opinion mining or PII detection

Use GPT-4o when:

• You need flexible analysis combined with other tasks
• The analysis requires understanding complex context
• You want natural language explanations alongside the analysis
• You’re already using GPT-4o for other features in the same app

🎬 Video walkthrough

🎬 Video coming soon

Building a Text Analysis App — AI-901 Module 17

Building a Text Analysis App — AI-901 Module 17

~14 min

Flashcards

Question

What Python package provides the Azure AI Language text analysis SDK?

Click or press Enter to reveal answer

Answer

azure-ai-textanalytics — provides TextAnalyticsClient with methods for sentiment analysis, entity recognition, key phrase extraction, and language detection.

Click to flip back

Question

What are the two approaches to building text analysis in Azure?

Click or press Enter to reveal answer

Answer

1) Azure AI Language SDK (dedicated NLP API, structured output, lower cost per transaction), or 2) GPT-4o via chat completions (flexible, natural language analysis, combined with other tasks).

Click to flip back

Question

What does the analyze_sentiment method return?

Click or press Enter to reveal answer

Answer

For each document: the overall sentiment (positive, negative, neutral, mixed) and confidence scores for each category (e.g., positive: 0.95, negative: 0.03, neutral: 0.02).

Click to flip back

Knowledge Check

Knowledge Check

DataFlow Corp processes 50,000 customer emails daily and needs to tag each with sentiment (positive/negative/neutral) in a consistent JSON format. Which approach is most appropriate?

AGPT-4o — for the highest quality analysisBAzure AI Language SDK — optimised for high-volume, structured text analysisCAzure Speech — to convert emails to audio firstDContent Understanding — to extract form fields from emails

Check Answer

Knowledge Check

MediSpark wants their app to read patient feedback, determine sentiment, extract the doctor's name and clinic location, and generate a personalised response. Which approach makes most sense?

AAzure AI Language SDK only — it handles all text analysis tasksBGPT-4o only — it can do analysis AND generate the response in one callCAzure AI Language for analysis, then GPT-4o for response generationDContent Understanding — it can process feedback forms

Check Answer

---

Next up: Multimodal: Responding to Speech — using AI to respond to spoken prompts.