Implement information extraction solutions 5.1 Build retrieval and grounding pipelines

A product index has a semantic configuration with title field productName and content…

Study guide skill: Configure semantic search, hybrid search, and vector search for grounding

Question B7-019

A product index has a semantic configuration with title field productName and content fields reviews, description, and specs, in that order. Long review text dominates semantic ranking, while the description field matters most, and a developer also wants to add the contentVector field as a content field. Which two statements about semantic configurations are correct? (Choose two.)

Select 2 answers.

  • A Only one content field is allowed, so reviews and specs must move to keyword fields
  • B Fields in a semantic configuration must be filterable rather than searchable
  • C Only searchable and retrievable string fields can be used, so contentVector can't be a content field ✓ Correct answer
  • D Add description as a second title field so it always takes priority
  • E List content fields in priority order, with description first, because lower-priority fields might get truncated ✓ Correct answer

Answer: C, E

A semantic configuration has one title field and any number of content and keyword fields, which you list in priority order because lower-priority fields might be truncated. Every assigned field must be searchable and retrievable and of a string type, and semantic ranking applies to nonvector string fields, so a vector field doesn't qualify.

Deep dive

How it works

A semantic configuration tells Azure AI Search's semantic ranker which fields to read, and in what order. It has three properties:

  • Title field: exactly one. You can specify only one title field.
  • Content fields: as many as you like, for the main descriptive text.
  • Keyword fields: as many as you like, for terms such as categories.

Order matters. List content and keyword fields in priority order, because lower-priority fields might be truncated. With reviews first, long review text fills the ranker's input and description gets cut. Reorder to description, then specs, then reviews.

Field rules. Every field you assign, in any property, must be:

  1. attributed as searchable and retrievable;
  2. a string type: Edm.String, Collection(Edm.String), or a string subfield of a complex type.

Semantic ranking applies to nonvector string fields, so contentVector can't be a content field. Vectors take part in vector or hybrid queries; the semantic ranker then reranks using the text fields.

The other options, and when you'd use them

  • Only one content field allowed: Only the title is limited to one. Content and keyword fields can be as many as you like.
  • Fields must be filterable: The requirement is searchable and retrievable string fields. Filterable isn't part of it.
  • Description as a second title field: You can specify only one title field. Put description first in the content list instead.

Exam tip

Semantic configuration: one title, many content and keyword fields in priority order (lower ones may be truncated), all searchable + retrievable strings, no vectors.

References

Last reviewed:

Ready to test yourself? The quiz asks this and every other AI-103 question, with instant feedback. Practice this in a quiz →

All Information extraction questions · AI-103 study guide

Thanks for your feedback!