What is document and conversation summarization (preview)?

Important

  • To use gated summarization features, e.g. document abstractive summarization, conversation issue and resolution summarization, and conversation narrative summarization with chapters, you must submit an online request and have it approved.
  • These gated summarization features are only available through Language resources in the following regions:
    • North Europe
    • East US
    • UK South
  • Conversation issue and resolution summarization is only available using:
    • REST API
    • Python
    • C#
  • Document abstractive summarization, and conversation narrative summarization with chapter is only available using:
    • REST API

Summarization is one of the features offered by Azure Cognitive Service for Language, a collection of machine learning and AI algorithms in the cloud for developing intelligent applications that involve written language. Use this article to learn more about this feature, and how to use it in your applications.

This documentation contains the following article types:

  • Quickstarts are getting-started instructions to guide you through making requests to the service.
  • How-to guides contain instructions for using the service in more specific or customized ways.

Document summarization uses natural language processing techniques to generate a summary for documents. There are two general approaches to automatic summarization, both of which are supported by the API: extractive and abstractive.

Extractive summarization extracts sentences that collectively represent the most important or relevant information within the original content. Abstractive summarization generates a summary with concise, coherent sentences or words which are not simply extract sentences from the original document. These features are designed to shorten content that could be considered too long to read.

Key features

There are two types of document summarization this API provides:

  • Extractive summarization: Produces a summary by extracting salient sentences within the document.
    • Multiple extracted sentences: These sentences collectively convey the main idea of the document. They’re original sentences extracted from the input document’s content.
    • Rank score: The rank score indicates how relevant a sentence is to a document's main topic. Document summarization ranks extracted sentences, and you can determine whether they're returned in the order they appear, or according to their rank.
    • Multiple returned sentences: Determine the maximum number of sentences to be returned. For example, if you request a three-sentence summary extractive summarization will return the three highest scored sentences.
    • Positional information: The start position and length of extracted sentences.
  • Abstractive summarization: Generates a summary that may not use the same words as those in the document, but captures the main idea.
    • Summary texts: Abstractive summarization returns a summary for each contextual input range within the document. A long document may be segmented so multiple groups of summary texts may be returned with their contextual input range.
    • Contextual input range: The range within the input document that was used to generate the summary text.

As an example, consider the following paragraph of text:

"At Microsoft, we have been on a quest to advance AI beyond existing techniques, by taking a more holistic, human-centric approach to learning and understanding. As Chief Technology Officer of Azure AI Cognitive Services, I have been working with a team of amazing scientists and engineers to turn this quest into a reality. In my role, I enjoy a unique perspective in viewing the relationship among three attributes of human cognition: monolingual text (X), audio or visual sensory signals, (Y) and multilingual (Z). At the intersection of all three, there’s magic—what we call XYZ-code as illustrated in Figure 1—a joint representation to create more powerful AI that can speak, hear, see, and understand humans better. We believe XYZ-code will enable us to fulfill our long-term vision: cross-domain transfer learning, spanning modalities and languages. The goal is to have pre-trained models that can jointly learn representations to support a broad range of downstream AI tasks, much in the way humans do today. Over the past five years, we have achieved human performance on benchmarks in conversational speech recognition, machine translation, conversational question answering, machine reading comprehension, and image captioning. These five breakthroughs provided us with strong signals toward our more ambitious aspiration to produce a leap in AI capabilities, achieving multi-sensory and multilingual learning that is closer in line with how humans learn and understand. I believe the joint XYZ-code is a foundational component of this aspiration, if grounded with external knowledge sources in the downstream AI tasks."

The document summarization API request is processed upon receipt of the request by creating a job for the API backend. If the job succeeded, the output of the API will be returned. The output will be available for retrieval for 24 hours. After this time, the output is purged. Due to multilingual and emoji support, the response may contain text offsets. See how to process offsets for more information.

Using the above example, the API might return the following summarized sentences:

Extractive summarization:

  • "At Microsoft, we have been on a quest to advance AI beyond existing techniques, by taking a more holistic, human-centric approach to learning and understanding."
  • "We believe XYZ-code will enable us to fulfill our long-term vision: cross-domain transfer learning, spanning modalities and languages."
  • "The goal is to have pre-trained models that can jointly learn representations to support a broad range of downstream AI tasks, much in the way humans do today."

Abstractive summarization:

  • "Microsoft is taking a more holistic, human-centric approach to learning and understanding. We believe XYZ-code will enable us to fulfill our long-term vision: cross-domain transfer learning, spanning modalities and languages. Over the past five years, we have achieved human performance on benchmarks in."

Get started with summarization

To use this feature, you submit raw unstructured text for analysis and handle the API output in your application. Analysis is performed as-is, with no additional customization to the model used on your data. There are two ways to use summarization:

Development option Description Links
Language Studio A web-based platform that enables you to try document summarization without needing writing code. Language Studio website
Quickstart: Use Language Studio
REST API or Client library (Azure SDK) Integrate document summarization into your applications using the REST API, or the client library available in a variety of languages. Quickstart: Use document summarization

Input requirements and service limits

  • Summarization takes raw unstructured text for analysis. See Data and service limits in the how-to guide for more information.
  • Summarization works with a variety of written languages. See language support for more information.

Reference documentation and code samples

As you use document summarization in your applications, see the following reference documentation and samples for Azure Cognitive Services for Language:

Development option / language Reference documentation Samples
REST API REST API documentation
C# C# documentation C# samples
Java Java documentation Java Samples
JavaScript JavaScript documentation JavaScript samples
Python Python documentation Python samples

Responsible AI

An AI system includes not only the technology, but also the people who will use it, the people who will be affected by it, and the environment in which it’s deployed. Read the transparency note for summarization to learn about responsible AI use and deployment in your systems. You can also see the following articles for more information: