Implement text analysis solutions 4.1 Apply language model text analysis
An engineering firm must tag proprietary part numbers and internal project code names in…
Study guide skill: Implement solutions to extract entities, topics, summaries, and structured JSON outputs by using generative prompting and Foundry Tools
Question B2-026
An engineering firm must tag proprietary part numbers and internal project code names in maintenance reports. The prebuilt entity categories don't include these. Which Azure Language capability fits?
Select one answer.
Answer: D
Prebuilt NER recognizes a preset list of entity categories such as people, places, organizations, and quantities. Custom NER lets you train a model to recognize specialized entities for your use case, such as part numbers and code names. PII detection targets personal data, and key phrase extraction doesn't classify entities into your own types.
Deep dive
How it works
Azure Language has two flavors of named entity recognition. Prebuilt NER identifies and categorizes entities in unstructured text, such as people, places, organizations and quantities, from a preset list of entity categories. Custom NER lets you train a model to recognize specialized entities specific to your use case. Proprietary part numbers and internal code names aren't in any preset list, so custom NER is the fit.
A custom NER project follows a loop: label data, train, evaluate and improve the model, then deploy it. The docs stress that the quality of your labeled data directly affects the model's accuracy, so label the new entity types carefully and consistently. You can build and label the model through a web platform in Microsoft Foundry. Typical scenarios include extracting domain-specific entities from unstructured text, such as contracts or financial documents, and enriching knowledge-mining pipelines for semantic search.
The other options, and when you'd use them
- Prebuilt named entity recognition: Use it when the categories you need are already in the preset list, such as people, places, organizations or quantities. It needs no training, but it can't learn your part-number format.
- PII detection with redaction: PII detection identifies, classifies and redacts sensitive data such as names, addresses, financial IDs and health identifiers. Use it for privacy, not for tagging engineering terms.
- Key phrase extraction: Key phrase extraction quickly identifies the main concepts in text. For example, it returns "food" and "wonderful staff" from a short review. It finds topics but doesn't assign them to entity types you define. Its Learn page currently carries a retirement notice, so check it before you start new work on it.
Exam tip
"Prebuilt categories don't cover our domain terms" means custom NER, trained on labeled examples. "Standard people, places and organizations" means prebuilt NER, and "remove personal data" means PII.
References
Last reviewed: