Last March we announced semantic search for Azure Cognitive Search, a feature that retrieves search results based on the semantic meaning of search terms (not just keywords). We enabled this by leveraging the state-of-the-art transformer-based language models that power the Microsoft Bing search engine. Since our initial announcement, we have seen a wide adoption from customers across industries such as retail, financial services, manufacturing, legal and professional services. These customers are leveraging the advanced capabilities of semantic search – semantic ranking, captions and answers (with semantic highlighting) – to light up rich search scenarios across several languages, all without needing to worry about training large scale AI models or managing expensive infrastructure to deploy them. During this period, Forrester also named us a leader in the 2021 Cognitive Search Wave, further underscoring the value of making Bing’s AI available to developers.
Today, we are excited to announce that we are making semantic search available to even more customers across an expanded set of regions – and customers can now enable it directly from the Azure portal. We are also releasing semantic configurations – an intuitive experience to help customers seamlessly set up semantic search for their datasets.
Semantic Configurations
Up until now, semantic search could be configured only by using the searchFields query parameter to specify an ordered list of fields. While this one-step configuration allowed for minimal effort, we received feedback from many users on how the experience was not very intuitive and required multiple iterations to arrive at the right order. To address this gap and further empower our users to effectively setup semantic search, we are happy to announce the release of semantic configurations. Using this feature, you can now easily bucket fields into categories such as document or catalog title, descriptive text or long-form content, and keywords or terms related to a document.
Semantic configurations are a part of your index definition, and you can easily create one in the Azure portal (see Figure 1). There is no need to re-index your content to create a semantic configuration or use semantic search.
Once you create a semantic configuration, you will specify it as part of your semantic query:
{
"search": "what is the incubation period for Covid-19?",
"queryType": "semantic",
"queryLanguage": "en-us",
"captions": "extractive",
"answers": "extractive",
"semanticConfiguration": "my-semantic-config"
}
Additionally, we have updated our underlying relevance models to enhance our semantic ranking capabilities to deliver even better search relevance to our customers using semantic configurations.
You can read more about the models behind semantic search in our earlier blog post.
Customer Momentum
As mentioned earlier, we have seen adoption across a broad spectrum of industries, spanning manufacturing, finance, legal, retail and more.
Howden, a global leader of manufacturing air and gas handling solutions, built their own search engine, Smart Records, to make industrial equipment records easily retrievable. Their solution harnesses Azure Cognitive Search’s orchestration pipeline to apply optical character recognition (OCR) to extract text from PDFs, images, or scanned documents. It then pulls out critical information like customer names, sales order numbers, and locations, adding them to the search index. They then incorporated semantic search, leveraging their deep neural network models to understand user intent and surface contextual results.
We have also adopted Semantic Search on our Microsoft Docs site to help visiting customers in the US more easily find information relevant to their search. Our documentation platform performs over 80,000 semantic search queries per day. Since implementing semantic search, Microsoft Docs has seen a 3% gain in clickthrough rate on their search results page.
Regional Expansion and Pricing
As a part of this release, we are also expanding semantic search availability to additional regions. semantic search is now available for customers who have search services in the standard tier (S1, S2 or S3) in the following regions: Australia East, East US, East US 2, North Central US, South Central US, West US, West US 2, North Europe, UK South and West Europe. We plan to continue expanding our support for more regions going forward.
Getting Started
With the preview now open to all customers, you can easily try out semantic search on your search service as long as you're in a supported tier and region:
Beyond signing up in the Azure portal, you can also:
- Read our documentation for how to get started.
- Check out our pricing page for semantic search regional pricing.
- Learn more about our models here: The science behind semantic search: How AI from Bing is powering Azure Cognitive Search - Microsoft ...
- Learn more about these changes in Episode 46 of the AI show
