Identify AI Workloads · Topic 1 of 3
AI Workloads and Text Analysis
AI-901 AI workloads: generative AI, agentic AI, text analysis, speech, vision and extraction, plus key phrases, entities, sentiment and summarisation.
- 5 min
- Generative AI creates new content from a prompt. Agentic AI uses tools to pursue a goal.
- Text analysis finds meaning in existing text: key phrases, entities, sentiment, summaries and language.
- Opinion mining links sentiment to specific aspects.
What it is
AI solutions fall into a handful of workloads. Each one takes a certain kind of input and produces a certain kind of result. Recognising the workload from a short scenario is a core exam skill.
Text analysis is the workload that finds meaning in existing text. It uses a set of standard techniques that the exam expects you to name.
How it works
The common workloads
| Workload | What it does | Example |
|---|---|---|
| Generative AI | Creates new text, code or images from a natural language prompt | Draft a customer reply, generate a product photo |
| Agentic AI | An agent that pursues a goal, deciding which tools to use and what to do next | A support agent that checks an order and issues a refund |
| Text analysis | Finds meaning in existing text | Detect the mood of reviews |
| Speech | Turns speech into text, and text into speech | Live captions, a voice assistant |
| Computer vision | Understands images and video | Describe a photo, read a sign |
| Image generation | Creates pictures from a description | Marketing images |
| Information extraction | Pulls structured data out of documents, images, audio and video | Fields from an invoice |
Other kinds of AI exist, such as anomaly detection, which finds unusual values in data, and forecasting, but they are not the focus of this exam.
Generative AI compared with analysis. Generative AI produces something new from a prompt. Text analysis, computer vision and information extraction analyse existing content and return what they found. That is the quickest way to tell them apart.
Agentic AI goes a step further than a single prompt and answer. An agent has instructions, a model and tools, and it decides how to reach a goal, for example by calling a function to look up data and then writing a reply.
Common text analysis techniques
| Technique | What it returns | Example |
|---|---|---|
| Key phrase extraction | The main talking points of a text | "slow shipping", "friendly staff" |
| Entity detection | Named things, such as people, places, organisations, dates and quantities | "Contoso", "Lisbon", "next Friday" |
| Sentiment analysis | Whether the text is positive, neutral, negative or mixed, with confidence scores | A review is negative |
| Summarisation | A shorter version of a longer text | A one paragraph summary of a report |
| Language detection | Which language the text is written in | A message is in Spanish |
Entity detection has useful variations. Entity linking connects a name to an entry in a knowledge base, so "Mercury" can be tied to the planet or to the element. PII detection finds personal data such as phone numbers and email addresses, so it can be hidden.
Sentiment analysis can also do opinion mining, which links sentiment to specific aspects. In the review "The food was unacceptable. The concierge was nice", opinion mining finds negative sentiment about the food and positive sentiment about the concierge.
Summarisation comes in two forms. Extractive summarisation picks the most important sentences from the original. Abstractive summarisation writes new sentences that capture the meaning.
Picking the workload from a scenario
| Scenario | Workload |
|---|---|
| A tool drafts a reply from a short instruction | Generative AI |
| Thousands of reviews need to be sorted by mood | Text analysis, sentiment |
| A voice assistant answers spoken questions | Speech with generative AI |
| A shop app reads the text on a product label in a photo | Computer vision |
| Fields are pulled from scanned invoices | Information extraction |
| A program decides which tools to call to finish a task | Agentic AI |
A worked scenario
A hotel chain collects thousands of guest reviews. It wants to know what guests talk about and whether they are happy. Key phrase extraction lists the topics, such as "breakfast" and "check in". Sentiment analysis with opinion mining says whether each topic is praised or criticised. Entity detection finds hotel names and cities. A summary of each week's reviews is then written for managers. All of this analyses existing text, so it is text analysis. If the chain also lets staff type "write a polite apology for a late check in", that request is generative AI, because it creates new text.
What to remember for the exam
- Generative AI creates new content from a prompt. Agentic AI uses tools to pursue a goal.
- Text analysis finds meaning in existing text: key phrases, entities, sentiment, summaries and language.
- Opinion mining links sentiment to specific aspects.
- Extractive summaries select sentences. Abstractive summaries write new ones.
- Entity linking ties a name to a knowledge base entry. PII detection finds personal data.
If the scenario creates something new, think generative AI. If it analyses something that already exists, think text analysis, vision, speech or extraction.
How the exam asks about this
- "Which workload produces new text, code or images in response to a prompt?" Generative AI.
- "Which technique lists the main talking points in a text?" Key phrase extraction.
- "Which technique reports whether feedback is positive or negative?" Sentiment analysis.
- "Which technique finds people, places and dates in text?" Entity detection.
Common mistakes
Calling sentiment analysis generative AI. It classifies existing text. It does not create new text.
Mixing up key phrases and entities. Key phrases are the main topics. Entities are named things.
Treating a summary as always extractive. Abstractive summaries write new wording.
Using anomaly detection for text mood. That workload looks for unusual values in data.
Check yourself
Which workload creates new content from a prompt?
Generative AI.
What does opinion mining add to sentiment analysis?
It links the sentiment to specific aspects mentioned in the text.
What is the difference between extractive and abstractive summarisation?
Extractive picks key sentences from the original. Abstractive writes new sentences.
Topic 1 of 3 in Identify AI Workloads
Test yourself on Identify AI Workloads
The full Microsoft AI-901 question bank has 511 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Microsoft. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.