Text Analysis With Azure Language in Foundry: Sentiment, Entities, Key Phrases (AI-901)

Text analysis turns unstructured text, such as reviews, emails, tickets and articles, into insight. For AI-901 you need to know the techniques and how to "build a lightweight application that includes text analysis" in Foundry.

There are two ways to do text analysis in Foundry:

1. Azure Language in Foundry Tools (formerly Azure AI Language): purpose-built, pre-trained features with consistent, structured output and confidence scores.
2. A generative model with a good prompt: flexible, handles unusual tasks, but output needs careful formatting and validation.

The exam can test both. Know when each fits.

Azure Language features

Build a text analysis app in Python

Install the client library:

pip install azure-ai-textanalytics

You need your Foundry (or Language) resource endpoint and a credential. A key is shown here for simplicity; in production use Entra ID or store the key in Azure Key Vault.

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

client = TextAnalyticsClient(
    endpoint=os.environ["LANGUAGE_ENDPOINT"],
    credential=AzureKeyCredential(os.environ["LANGUAGE_KEY"]),
)

reviews = [
    "The hotel in Lahore was spotless and the staff were wonderful.",
    "Check-in took an hour and the Wi-Fi never worked. Call me on 0300-1234567.",
]

# Language
for doc in client.detect_language(reviews):
    print("Language:", doc.primary_language.name)

# Sentiment
for doc in client.analyze_sentiment(reviews):
    print("Sentiment:", doc.sentiment, doc.confidence_scores)

# Key phrases
for doc in client.extract_key_phrases(reviews):
    print("Key phrases:", doc.key_phrases)

# Entities
for doc in client.recognize_entities(reviews):
    for e in doc.entities:
        print("Entity:", e.text, "|", e.category, "|", round(e.confidence_score, 2))

# PII, with redacted text
for doc in client.recognize_pii_entities(reviews):
    print("Redacted:", doc.redacted_text)

Notice the pattern: create a client with endpoint + credential → call a method with a list of documents → loop over the results. Each result has an is_error flag you should check in real code.

Expected kinds of output:

• Review 1: sentiment positive; key phrases such as hotel, staff; entity Lahore (Location).
• Review 2: sentiment negative; key phrases such as check-in, Wi-Fi; PII detection finds the phone number and returns the text with it masked.

Summarization

Summaries run as long-running operations:

poller = client.begin_abstract_summary([long_article])
for result in poller.result():
    for summary in result.summaries:
        print(summary.text)

begin_extract_summary does extractive summarization the same way.

Doing it with a generative model instead

response = openai.responses.create(
    model="gpt-5-mini",
    instructions="Analyse customer reviews. Return JSON with: sentiment "
                 "(positive/negative/mixed), key_phrases (array) and summary (one sentence).",
    input=review_text,
)

Techniques behind the features

Fundamentals questions sometimes ask about classic NLP ideas:

• Tokenization: splitting text into words or sub-words.
• Stop-word removal: dropping common words such as "the" and "is".
• Stemming / lemmatization: reducing words to a base form (running → run).
• TF-IDF: scoring words that are common in one document but rare across all documents, a simple form of keyword extraction.
• Embeddings: vectors that capture meaning, used by modern models for sentiment, similarity and search.

Key takeaways

• Know each Language feature and its output: language, sentiment and opinions, key phrases, entities, PII, summaries.
• SDK pattern: TextAnalyticsClient(endpoint, credential) → analyze_sentiment, extract_key_phrases, recognize_entities, recognize_pii_entities, detect_language.
• Extractive summaries reuse sentences; abstractive summaries write new ones.
• Use Language for consistent, scalable analysis; use a generative model for flexible or unusual tasks.

Next lesson: speech AI on Azure.