Text Analysis: Keywords, Entities & Sentiment

What is text analysis?

Simple explanation

Text analysis is when AI reads text and tells you what it’s about, how the writer feels, and what the key facts are.

Imagine you have 10,000 customer reviews. You can’t read them all. Text analysis AI reads every single one and tells you: “85% are positive, people love the delivery speed, the main complaint is about packaging, and the most mentioned product is the wireless earbuds.”

It doesn’t create new content — it understands existing text.

The four key techniques

1. Keyword extraction

What it does: Identifies the most important terms and phrases in a document.

Example input: “Microsoft announced a new AI platform called Foundry that combines model deployment, agent creation, and enterprise governance into a single Azure service.”

Keywords extracted: Microsoft, AI platform, Foundry, model deployment, agent creation, enterprise governance, Azure

DataFlow Corp scenario: DataFlow processes 50,000 support tickets per month. Keyword extraction automatically tags each ticket with relevant topics (billing, outage, performance) for routing and reporting.

2. Entity detection (Named Entity Recognition)

What it does: Identifies and classifies specific entities in text — people, places, organisations, dates, amounts.

Entity Type	Example	Category
Person	”Dr. Sarah Chen”	Person
Organisation	”MediSpark”	Organisation
Location	”Auckland, New Zealand”	Location
Date	”April 21, 2026”	DateTime
Money	”$4.5 million”	Quantity
Email	”support@medispark.com”	Contact

GreenLeaf scenario: GreenLeaf scans hundreds of supplier contracts. Entity detection automatically extracts supplier names, contract dates, payment amounts, and delivery locations.

3. Sentiment analysis

What it does: Determines whether text expresses positive, negative, neutral, or mixed feelings.

Text	Sentiment	Confidence
”The product is amazing, best purchase ever!”	Positive	98%
“Delivery was slow and the box was damaged”	Negative	91%
“The order arrived on Tuesday”	Neutral	85%
“Love the quality but hate the price”	Mixed	76%

MediSpark scenario: MediSpark analyses patient feedback surveys. Sentiment analysis flags negative responses about wait times for immediate review, while tracking overall satisfaction trends.

Opinion mining — beyond simple sentiment

Opinion mining goes deeper than sentiment analysis by identifying what specific aspect the sentiment is about:

“The food was amazing” → food: positive
“The service was terrible” → service: negative
“The location is convenient but the parking is awful” → location: positive, parking: negative

This is called aspect-based sentiment analysis and is available in Azure AI Language.

4. Summarisation

What it does: Condenses long text into a shorter version while preserving key information.

Two types:

Extractive summarisation — pulls the most important sentences directly from the source
Abstractive summarisation — generates new sentences that capture the meaning (uses generative AI)

Extractive vs abstractive summarisation
Feature	Extractive	Abstractive
How it works	Selects key sentences from the original text	Generates new sentences that summarise the meaning
Accuracy	Words are exactly from the source	May rephrase — risk of minor inaccuracies
Natural flow	Can feel choppy (sentences from different parts)	Reads more naturally, like a human summary
Technology	Traditional NLP models	Generative AI (LLMs)
Best for	Legal documents, contracts (exact wording matters)	Meeting notes, news articles, reports

Where text analysis fits in Azure

Service	Capability	Type
Azure AI Language (Foundry Tools)	Keyword extraction, entity detection, sentiment analysis, summarisation	Dedicated NLP service
GPT-4o / multimodal models	All text analysis tasks + generation	General-purpose LLM

Exam tip: Azure AI Language is a dedicated service optimised for text analysis. But GPT-4o can also do text analysis as part of a broader conversation. Know that both options exist.

🎬 Video walkthrough

🎬 Video coming soon

Text Analysis Techniques — AI-901 Module 7

~14 min

Flashcards

Question

What are the four text analysis techniques tested in AI-901?

Click or press Enter to reveal answer

Answer

Keyword extraction (important terms), entity detection (people, places, dates), sentiment analysis (positive/negative/neutral), and summarisation (condensing text).

Click to flip back

Question

What is Named Entity Recognition (NER)?

Click or press Enter to reveal answer

Answer

A text analysis technique that identifies and classifies specific entities in text — people, organisations, locations, dates, amounts, emails — and labels each with its category.

Click to flip back

Question

What is the difference between extractive and abstractive summarisation?

Click or press Enter to reveal answer

Answer

Extractive pulls the most important sentences directly from the source text. Abstractive generates new sentences that capture the meaning using generative AI. Extractive is more accurate; abstractive reads more naturally.

Click to flip back

Question

What is opinion mining?

Click or press Enter to reveal answer

Answer

An advanced form of sentiment analysis that identifies what specific aspect the sentiment is about. Example: 'Great food but terrible service' → food: positive, service: negative.

Click to flip back

Knowledge Check

DataFlow Corp wants to automatically categorise 50,000 monthly support tickets by topic. Each ticket should be tagged with relevant topics like 'billing', 'outage', or 'account access'. Which text analysis technique is most appropriate?

Knowledge Check

MediSpark analyses patient feedback: 'Dr. Chen was incredibly kind, but the 3-hour wait was unacceptable. The facility at 42 Queen Street was clean.' Which text analysis technique identifies 'Dr. Chen', '42 Queen Street', and '3-hour' as specific entities?

Next up: Speech: Recognition & Synthesis — how AI converts spoken words to text and text to natural-sounding speech.