Dataset

The canonical title of the dataset. This should be the full, official name as it appears in the accompanying paper or repository.

• Best Practice: Avoid truncating the title for display purposes; use the complete, authoritative string.
• Example: "The Pile: An 800GB Dataset of Diverse Text for Language Modeling"
• Entity Linking: This string is often the primary signal used to disambiguate the dataset in knowledge graphs.

A detailed summary of the dataset's contents, collection methodology, and intended use cases. This field is critical for discoverability in semantic search.

• Content: Describe the modality (text, image, tabular), the volume, and the annotation process.
• AI Context: Foundation model crawlers parse this field to understand the dataset's domain for training data attribution.
• Example: "A multi-terabyte, English-language corpus derived from 22 diverse high-quality subsets..."

The direct landing page where a human can access the dataset, typically a hosting platform like Hugging Face, Zenodo, or a dedicated institutional portal.

• Canonicality: This must be the single, stable URL. Avoid linking to transient cloud storage buckets.
• Relationship: Distinct from contentUrl (which points to the direct download file) and sameAs (which links to external knowledge bases).

The Organization or Person responsible for generating the dataset. Defining the creator establishes provenance and authority.

• Structured Value: Use a nested Organization or Person type with its own name and sameAs properties rather than a plain text string.
• Citation Impact: AI models use this linkage to attribute credit and assess the reliability of the source based on the creator's established reputation.

An array of DataDownload objects detailing how to obtain the dataset in specific formats. This separates the logical dataset from its physical instances.

• Properties: Each distribution should specify encodingFormat (e.g., "text/csv", "application/json"), contentUrl (direct download), and contentSize.
• Utility: Enables automated agents to select the correct file format without manual inspection.

A URL or CreativeWork defining the legal terms governing the dataset's use. This is a non-negotiable signal for compliance filtering in enterprise AI pipelines.

• Standardization: Use a canonical URL from SPDX or Creative Commons (e.g., https://spdx.org/licenses/CC-BY-4.0).
• Risk Mitigation: AI governance tools scan this property to prevent the ingestion of restrictively licensed data into commercial models.

A Schema.org type that represents a curated collection of datasets. Use DataCatalog to group related datasets under a single organizational structure, providing a landing page for discoverability. Key properties include:

• dataset: Links to individual Dataset items within the catalog.
• provider: The organization making the catalog available.
• keywords: Controlled vocabulary terms for topical classification.

DataCatalog is essential for large-scale data publishers, such as government open data portals or enterprise data marketplaces, to enable AI crawlers to navigate hierarchical data inventories.

A Schema.org type representing a specific downloadable file associated with a Dataset. It describes the physical representation of data, distinct from the abstract Dataset entity. Critical properties include:

• encodingFormat: The MIME type or file format (e.g., text/csv, application/json).
• contentSize: The file size in bytes.
• sha256: A cryptographic hash for integrity verification.

Defining DataDownload nodes allows AI systems to directly access and validate raw data files while maintaining clear provenance links to the parent Dataset.

A Schema.org type used to define the specific metric or variable measured within a Dataset. It disambiguates what each column or data point represents. Key properties:

• measuredProperty: The phenomenon being quantified.
• unitCode: The unit of measurement using UN/CEFACT codes.
• populationType: The group the variable applies to.

StatisticalVariable markup transforms raw tabular data into semantically rich, machine-readable statistical observations that AI models can reason over without human interpretation.

A Schema.org type that describes a quantity of currency, often used in conjunction with Dataset to specify licensing costs or access fees. It combines a numerical value with a currency code. Key properties:

• value: The numerical amount.
• currency: The ISO 4217 currency code (e.g., USD, EUR).

When a Dataset has commercial licensing terms, embedding MonetaryAmount within an Offer provides AI agents with structured pricing data for automated procurement and compliance checks.

A Schema.org type for marking up a word, name, or phrase with its formal definition. Within a Dataset context, DefinedTerm is used to create a codebook or data dictionary. Key properties:

• termCode: A unique identifier for the term.
• description: The full definition.
• inDefinedTermSet: Links to the parent DefinedTermSet.

Using DefinedTerm alongside Dataset ensures that categorical codes and domain-specific jargon are explicitly defined, eliminating ambiguity for LLMs performing retrieval-augmented generation.

A Schema.org type representing a property-value pair, used to express custom attributes of a Dataset that don't fit standard properties. It provides an extensibility mechanism. Key properties:

• propertyID: A URL or IRI identifying the property.
• value: The data value for the property.
• unitCode: Optional unit of measurement.

PropertyValue allows publishers to encode domain-specific metadata—such as sensor calibration parameters or collection methodology codes—while maintaining full Schema.org compliance.