Semantic search in Azure Cognitive Search


Semantic search is in public preview under supplemental terms of use. It's available through the Azure portal, preview REST API, and beta SDKs. These features are billable. For more information about, see Availability and pricing.

Semantic search is a collection of query-related capabilities that bring semantic relevance and language understanding to search results. This article is a high-level introduction to semantic search all-up, with descriptions of each feature and how they work collectively. The embedded video describes the technology, and the section at the end covers availability and pricing.

Semantic search is a premium feature. We recommend this article for background, but if you'd rather get started, follow these steps:

Semantic search is collection of features that improve the quality of search results. When enabled on your search service, it extends the query execution pipeline in two ways. First, it adds secondary ranking over an initial result set, promoting the most semantically relevant results to the top of the list. Second, it extracts and returns captions and answers in the response, which you can render on a search page to improve the user's search experience.

Feature Description
Semantic re-ranking Uses the context or semantic meaning to compute a new relevance score over existing results.
Semantic captions and highlights Extracts sentences and phrases from a document that best summarize the content, with highlights over key passages for easy scanning. Captions that summarize a result are useful when individual content fields are too dense for the results page. Highlighted text elevates the most relevant terms and phrases so that users can quickly determine why a match was considered relevant.
Semantic answers An optional and additional substructure returned from a semantic query. It provides a direct answer to a query that looks like a question. It requires that a document have text with the characteristics of an answer.
Spell check Corrects typos before the query terms reach the search engine.

How semantic ranking works

Semantic ranking looks for context and relatedness among terms, elevating matches that make more sense given the query. Language understanding finds summarizations or captions and answers within your content and includes them in the response, which can then be rendered on a search results page for a more productive search experience.

State-of-the-art pretrained models are used for summarization and ranking. To maintain the fast performance that users expect from search, semantic summarization and ranking are applied to just the top 50 results, as scored by the default similarity scoring algorithm. Using those results as the document corpus, semantic ranking re-scores those results based on the semantic strength of the match.

The underlying technology is from Bing and Microsoft Research, and integrated into the Cognitive Search infrastructure as an add-on feature. For more information about the research and AI investments backing semantic search, see How AI from Bing is powering Azure Cognitive Search (Microsoft Research Blog).

The following video provides an overview of the capabilities.

Order of operations

Components of semantic search extend the existing query execution pipeline in both directions. If you enable spelling correction, the speller corrects typos at query onset, before terms reach the search engine.

Semantic components in query execution

Query execution proceeds as usual, with term parsing, analysis, and scans over the inverted indexes. The engine retrieves documents using token matching, and scores the results using the default similarity scoring algorithm. Scores are calculated based on the degree of linguistic similarity between query terms and matching terms in the index. If you defined them, scoring profiles are also applied at this stage. Results are then passed to the semantic search subsystem.

In the preparation step, the document corpus returned from the initial result set is analyzed at the sentence and paragraph level to find passages that summarize each document. In contrast with keyword search, this step uses machine reading and comprehension to evaluate the content. Through this stage of content processing, a semantic query returns captions and answers. To formulate them, semantic search uses language representation to extract and highlight key passages that best summarize a result. If the search query is a question - and answers are requested - the response will also include a text passage that best answers the question, as expressed by the search query.

For both captions and answers, existing text is used in the formulation. The semantic models do not compose new sentences or phrases from the available content, nor does it apply logic to arrive at new conclusions. In short, the system will never return content that doesn't already exist.

Results are then re-scored based on the conceptual similarity of query terms.

To use semantic capabilities in queries, you'll need to make small modifications to the search request, but no extra configuration or reindexing is required.

Semantic capabilities and limitations

Semantic search is a newer technology so it's important to set expectations about what it can and cannot do. What it can do is improve the quality of search by:

  • Promoting matches that are semantically closer to the intent of original query.

  • Finding strings in each result that can be used as captions, and potentially answers, that can be rendered on a search results page.

What it can't do is rerun the query over the entire corpus to find semantically relevant results. Semantic search re-ranks the existing result set, consisting of the top 50 results as scored by the default ranking algorithm. Furthermore, semantic search cannot create new information or strings. Captions and answers are extracted verbatim from your content so if the results do not include answer-like text, the language models will not produce one.

Although semantic search is not beneficial in every scenario, certain content can benefit significantly from its capabilities. The language models in semantic search work best on searchable content that is information-rich and structured as prose. A knowledge base, online documentation, or documents that contain descriptive content see the most gains from semantic search capabilities.

Availability and pricing

Semantic search is available through sign-up registration. There is one sign-up for both semantic search and spell check.

Feature Tier Region Sign up Pricing
Semantic search (captions, highlights, answers) Standard tier (S1, S2, S3) North Central US, West US, West US 2, East US 2, North Europe, West Europe Required Cognitive Search pricing page
Spell check Any North Central US, West US, West US 2, East US 2, North Europe, West Europe Required None (free)

You can use spell check without semantic search, free of charge. Charges for semantic search are levied when query requests include queryType=semantic and the search string is not empty (for example, search=pet friendly hotels in new york. Empty search (queries where search=*) are not charged, even if queryType is set to semantic.

Only a search service that has the feature enabled can incur charges. However, if you want full protection against accidental usage, set the disabled option.

  • Management REST API version 2021-04-01-Preview provides this option

  • Owner or Contributor permissions are required to disable features

      "location": "{{region}}",
      "sku": {
        "name": "standard"
      "properties": {
        "semanticSearch": "disabled"

To re-enable semantic search, rerun the above request, setting "semanticSearch" to either "free" (default) or "standard".


Management REST API calls are authenticated through Azure Active Directory. For guidance on setting up a security principle and a request, see this blog post Azure REST APIs with Postman (2021). The previous example was tested using the instructions and Postman collection provided in the blog post.

Next steps

Sign-up for the preview on a search service that meets the tier and regional requirements noted in the previous section.

It can take up to two business days to process the request. Once your service is ready, create a semantic query to evaluate its performance on your content.