AI-900 Practice Question: Describe features of Natural Language Processing workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Text Analytics for Health

  Why this is correct

  Text Analytics for Health is a prebuilt Azure AI Language capability specifically designed to extract medical entities from unstructured clinical text, such as diagnoses, medications, procedures, and symptoms, using a healthcare-specific ontology (like UMLS and SNOMED CT). It requires no custom training or labeled data, making it ideal for scenarios where you simply need to mine existing patient notes. Additionally, it can extract relationships between entities (e.g., drug dosage or condition status) and assertion information like whether a condition was present, absent, or conditional. This is precisely why it is the correct answer for a hospital that needs to derive structured medical insights from free-text notes without building a custom model.

• ✗

  Custom Named Entity Recognition

  Why it's wrong here

  Custom Named Entity Recognition (NER) is a feature in Azure AI Language that allows you to build a bespoke model to identify entities that are specific to your domain, but it requires you to provide labeled training data and then train and evaluate the model, which the scenario explicitly states you want to avoid. Even if you were willing to train, you would need to annotate many patient notes manually, and then the model would still need to be retrained periodically to maintain accuracy. For extracting standard medical entities, a prebuilt healthcare model like Text Analytics for Health already exists, making custom NER an unnecessary, labor-intensive overshoot. Therefore, this option is incorrect because the user explicitly wants a no-training solution.

  When this WOULD be correct

  A hospital needs to extract unique entity types (e.g., internal department codes, local abbreviations) from patient notes that are not covered by pre-built models. The question would state that labeled training data is available and a custom solution is required.

• ✗

  Key Phrase Extraction

  Why it's wrong here

  Key Phrase Extraction is a prebuilt Azure AI Language feature that returns a list of the most salient phrases in a document based on statistical measures like term frequency and position, but it is domain-agnostic and not aware of medical terminology or clinical ontologies. For example, it might list 'chest pain' or 'the patient' as key phrases, but it cannot categorize those entities into a diagnosis, symptom, or medication, nor can it link them to standardized codes like ICD-10 or SNOMED CT. This makes it insufficient for clinical use cases where structured, coded medical entities are required. Unlike Text Analytics for Health, Key Phrase Extraction does not provide relation or assertion data, so it is the wrong tool for extracting meaningful medical information from patient notes.

  When this WOULD be correct

  A question asks: 'Which Azure AI Language feature extracts the most important points or topics from a document without any customization?' Key Phrase Extraction would be correct for summarizing main themes in general text.

• ✗

  Sentiment Analysis

  Why it's wrong here

  Sentiment Analysis in Azure AI Language is a prebuilt model that determines the emotional tone of a piece of text, classifying it as positive, negative, neutral, or mixed. It is often used for analyzing product reviews, social media posts, or customer feedback, but it does not extract domain-specific entities like medical conditions or medications. Running sentiment analysis on patient notes might tell you whether the overall text sounds positive or negative, but it will not structure the notes into meaningful clinical data such as diagnoses or procedures. Consequently, it fails to meet the hospital's need to extract relevant medical information from unstructured free text.

  When this WOULD be correct

  A social media monitoring company wants to automatically classify customer feedback about a new product as positive, negative, or neutral. They need a pre-built feature that does not require custom training. Sentiment Analysis would be the correct answer.

Option-by-option analysis

Why each answer is right or wrong

Understanding why wrong answers are wrong — and when they would be correct — is what separates a 750 score from a 900. The AI-900 exam frequently reuses these exact scenarios with slightly different constraints.

Analysis generated from the official AI-900blueprint and verified against question context. The “when correct” sections are what AI assistants cite when candidates ask “what’s the difference between these options?”

Go deeper

Related to this question