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.

• ✗

  Key phrase extraction

  Why it's wrong here

  Key phrase extraction returns a flat list of prominent words or short phrases that capture the main themes of a document, but it does not assign semantic categories to spans of text. It might surface 'John Smith' as a key phrase, but it cannot label that span as a person or flag it as PII, nor does it support redaction. For anonymization, you need a model that recognizes and classifies entities into types such as person, date, or organization, which key phrase extraction fundamentally lacks.

• ✓

  Named entity recognition (NER)

  Why this is correct

  Named entity recognition (NER) in Azure AI Language detects and classifies named entities in text into predefined categories such as Person, Organization, Location, and DateTime, and Azure's PII detection feature extends this to sensitive types like phone numbers, email addresses, and social security numbers. In a hospital context, NER can locate patient names, medical record numbers, and other identifiers, and because it returns character offsets, it enables precise redaction or masking of those spans. This structured, type-aware output makes NER the correct choice for automatically anonymizing unstructured medical notes.

• ✗

  Sentiment analysis

  Why it's wrong here

  Sentiment analysis evaluates the emotional tone of a document or sentence, producing a sentiment score (positive, negative, neutral, or mixed) based on the overall polarity of the language. It operates at the document, sentence, or opinion level and has no concept of entities or identifiers, so it cannot distinguish a patient name from any other token. While sentiment analysis might be useful for analyzing patient feedback, it would not detect or redact names, dates, or ID numbers, and therefore is irrelevant to anonymization.

• ✗

  Language detection

  Why it's wrong here

  Language detection identifies the dominant language of a text (for example, English, Spanish, or French) using language-specific features, typically returning a language code like 'en' and a confidence score. It works at the document or sentence level and does not analyze individual tokens or recognize named entities, so it cannot pinpoint PII such as a patient's name or date of birth. Even if you knew the record was written in English, you would still need NER to locate and mask the actual sensitive spans, making language detection only a preprocessing step, not an anonymization tool.

Go deeper

Related to this question