Azure Content Understanding in Foundry Tools terminology

Term	Description
Add-ons	Added features that enhance content extraction results, such as layout elements, barcodes, and figures in documents.
Analyzer	A component that processes and extracts content and structured fields from files. Content Understanding offers prebuilt analyzers for common scenarios and supports creating custom analyzers.
Analyzer result	The output generated by an analyzer after processing input data. It typically includes extracted content in Markdown, extracted fields, and optional modality-specific details.
Base analyzer	A foundational analyzer specific to a content type (document, image, audio, video, call center) that serves as a parent when creating custom analyzers. Use the baseAnalyzerId property to inherit from a base analyzer.
Categorization	A feature that classifies content into predefined categories and optionally routes it to specialized analyzers for further processing. Supports up to 200 categories within a single analyzer. Learn more about categorization.
Confidence score	Content Understanding provides confidence scores from 0 to 1 to estimate the reliability of the results. High scores indicate accurate data extraction, enabling straight-through processing in automation workflows. Learn more about confidence scores and field extraction.
Content extraction	The foundational processing step that transforms unstructured input into standardized format. Includes optical character recognition (OCR) for documents, speech-to-text for audio/video, layout detection, and structural element identification.
Contextualization	Content Understanding's processing layer that prepares context for generative models and post-processes their output into structured results. Includes output normalization, source grounding calculation, confidence score computation, and context engineering. Learn more in the Pricing explainer.
Domain-specific analyzer	A prebuilt analyzer preconfigured for common document categories in popular industries, such as invoices, tax forms, mortgage documents, and identity documents. See the complete list of domain-specific analyzers.
Field schema	A formal description of the fields to extract from the input. It specifies the name, description, value type, generation method, and more for each field. Learn more about field schema limits.
Fields	List of structured key-value pairs derived from the content, as defined by the field schema. Learn more about supported field value types.
File	Any type of data, including text, documents, images, videos, and audio.
File type	The MIME type of a file, such as text/plain, application/pdf, image/jpeg, audio/wav, and video/mp4. Generic categories like document refer to all corresponding MIME types supported by the service. See supported file formats.
Generation method	The process of determining the extracted value of a specified field. Content Understanding supports: • Extract: Directly extract values from the input content, such as dates from receipts or item details from invoices. • Classify: Classify content into predefined categories, such as call sentiment or chart type. • Generate: Generate values from input data, such as summarizing an audio conversation or generating scene descriptions from videos. Learn more about generation methods and best practices.
Grounding source	The specific regions in content where a value was extracted or generated. Available for both extract and generate methods when enabled. Learn more about confidence scores and source grounding.
Knowledge base	A collection of labeled training examples used to improve custom analyzer accuracy for domain-specific scenarios. Training examples are embedded and retrieved during analysis to guide the model. Learn more about creating custom analyzers with training examples.
Prebuilt analyzer	Ready-to-use analyzers provided by Content Understanding for common scenarios, including content extraction (read, layout), RAG workflows (documentAnalyzer, videoAnalyzer), and domain-specific tasks (invoice, tax forms, identity documents). See the complete list of prebuilt analyzers.
Processing Location	An API request parameter that defines the geographic region where Foundry Tools analyzes your data. You can choose from three options: geography, dataZone, and global to control where processing occurs. This setting helps meet data residency requirements and optimize performance or scalability based on your needs. Learn more about region support.
RAG analyzer	A prebuilt analyzer optimized for retrieval-augmented generation scenarios that extracts content with layout as markdown and performs semantic analysis to enhance retrieval quality. See RAG analyzers and the RAG solution tutorial.
Segmentation	The process of dividing documents or videos into logical sections for targeted processing. Configured using the segmentationMode property in the analyzer schema. Learn more about categorization for documents and categorization for videos.
Span	A reference indicating the location of an element (for example, field, word) within the extracted Markdown content. A character offset and length represent a span. Different programming languages use various character encodings, which can affect the exact offset and length values for Unicode text. To avoid confusion, spans are only returned if the desired encoding is explicitly specified in the request. Some elements can map to multiple spans if they aren't contiguous in the markdown (for example, page). Learn more about document elements and spans.