This page describes the RDF dump and export format produced by Wikidata and used for export and indexing purposes. Note that while it is close to the format used by the Wikidata Toolkit , it is not the same code and not the same format. While we strive to keep divergence to a minimum, there may be differences and one should use documentation only for the format that is actually being consumed.

This document describes the RDF dump as can be downloaded from Wikimedia dump source , and while it can be used to create queries for Wikidata query service, the service can have small differences in how the data there look like. See the chapter for the full list.

The canonical URI of the Wikibase RDF ontology is < http://wikiba.se/ontology >. The current version can be found at < http://wikiba.se/ontology-1.0.owl >.

Changes to the RDF mapping are subject to the Stable Interface Policy .

The RDF format is a binding for on the Wikibase data model and represents an export format for it. That means, in particular, that if/when the data model changes, the export format will be changed accordingly. This document will be updated for such changes. The following description assumes familiarity with the data model and the terminology used.

This RDF binding is based on the one designed for the Wikidata Toolkit by Denny Vrandecic and Markus Krötzsch, see http://korrekt.org/papers/Wikidata-RDF-export-2014.pdf .

The following description uses prefixes to describe the IRIs of the RDF resources mentioned. See the for the full description. All examples below are given in Turtle format.

The version of the data model is specified by schema:softwareVersion predicate of the schema:Dataset node, which is either dump node for the dump or entity data node ( wdata: ) for single entity page. Released versions:

For the RDF dump, there is the header node wikibase:Dump containing information about the license, the software version of the generator and the date the data was produced. In single-entity export, this data is attached to the data node (see below).

Example header:

The entity is described in two nodes - data node and entity node. For entity Q1, data node is wdata:Q1 and entity node is wd:Q1 .

Data node describes the metadata about the entity record in the Wikibase - i.e. data which are not part of the entity's information but instead describe the status of the entity inside Wikibase. It has type of schema:Dataset and contains the following metadata:

Example:

Entity node describes the actual entity data and has type wikibase:Item or wikibase:Property depending on the kind of entity. Other entity types can be introduced in the future.

Entity description includes the following:

Example of the entity definition:

Entity node can also carry additional information about the entity, such as number of links or statements. The data is sourced from page properties and can be specified in config file. For example:

wikibase:statements specifies how many statements this entity has, and wikibase:sitelinks specifies the number of sitelinks . Additional statements can be introduced in the future.

Entities that represent items have the common entity data as described above, plus can have attached to them, as described below.

Entities that represent properties additionally feature the property type using wikibase:propertyType predicate. The object of the predicate is the property type described in below, with wikibase: prefix and each word capitalized, with no separators. I.e., wikibase-item becomes wikibase:WikibaseItem .

Each property is also linked to the that are derived from it. Example:

The property predicates also have type definitions:

The type depends on the type of the original property - whether its is literal ( DatatypeProperty ) or IRI ( ObjectProperty ). However, p: , psv: , pqv: and prv: predicates would always be owl:ObjectProperty .

Note that wdno:P22 mentioned above is not a predicate, unlike others, but a class. See the full description of it in section.

Please see full description at Lexeme RDF mapping .

Lexemes are represented according to Lexeme RDF mapping . Example:

Please see full description at MediaInfo RDF mapping .

MediaInfo entities are represented according to MediaInfo RDF mapping . Example:

This example demonstrates the MediaInfo data on wikimedia commons when used in federation with wikidata.

The RDF format represents statements in two forms - truthy and full statements.

Truthy statements represent statements that have the best non-deprecated rank for given property. Namely, if there is a preferred statement for property P2, then only preferred statements for P2 will be considered truthy. Otherwise, all normal-rank statements for P2 are considered truthy.

Truthy statement predicates have prefix wdt: with the property name (e.g. wdt:P2 ) and the object is the simple value (see below) for the statement. The qualifiers are ignored.

If the value has simple value normalization (currently true only for external ID), normalized value is listed under wdtn: prefix, e.g. wdtn:P2 .

Full statements represent all data about the statement in the system. Full statement is represented as separate node, with prefix wds: with the id of the statement (e.g. wds:Q3-4cc1f2d1-490e-c9c7-4560-46c3cce05bb7 ). There is no guaranteed format or meaning to the statement id.

The statements are linked to the entity with the predicate with prefix p: and the name of the property (e.g. p:P2 ).

Warning:

It is possible that a statement refers to a property or item that no longer exists. Therefore you should not assumed the predicates and objects are always defined.

The statement node represents a single statement about the entity. It has type wikibase:Statement . The statement can contain the rank, the simple value (see below) of the statement, the link to the full value, the qualifiers and the references.

The statement rank is represented by the predicate wikibase:rank and the object being one of: wikibase:NormalRank , wikibase:PreferredRank , wikibase:DeprecatedRank .

The statement that has the best rank for the property (i.e., preferred if there are any preferred statements in the property, otherwise normal) is also has type of wikibase:BestRank .

The simple value is represented by the predicate with prefix ps: and the name of the property (e.g. ps:P2 ) and the object being the simple value.

The full value (if required by the type) is represented by the predicate with prefix psv: (e.g. psv:P2 ) and the object being the full value node.

The statement always has no more than one value, but can have multiple qualifiers and references.

The qualifiers are represented by predicates with prefix pq: and the name of the property (e.g. pq:P2 ) and the object being the simple value of the qualifier.

The full value (if required by the type) is represented by the predicate with prefix pqv: (e.g. pqv:P2 ) and the object being the full value node.

References are represented by the predicate prov:wasDerivedFrom with the object being the reference node (see below).

Example of the statement:

The service also may contain data for Wikibase Quality Constraints violations, via wikibase:hasViolationForConstraint predicate. The predicate links violating statement to the statement describing the constraint, e.g.:

Note that constraint violations are loaded from constraint cache and are not guaranteed to be up-to-date or present for all items (which means if you find no constraint violation statements for an item, that doesn't mean it doesn't have any - check Wikibase Quality tools for more up-to-date information).

References represent provenance information about statements.

Reference is represented as node, with prefix wdref: and the local name being the hash derived from the reference contents (e.g. wdref:d95dde070543a0e0115c8d5061fce6754bb82280  ). The exact value of the hash is not guaranteed beyond the fact that the same references (i.e. ones with identical content) will generate the same hash, and the different one will generate the different one. The same reference (i.e. reference having the same properties with the same values) will be usually represented with single node, though duplicate reference nodes are possible in the data.

The type of the node is a wikibase:Reference .

The reference values are represented the same as statement values, with simple values using predicates with pr: prefix (e.g. pr:P2 ) and full values with prefix prv: (e.g. prv:P2 ) and the object being the full value node. Unlike statements, references can have any number of values.

Example of the reference node:

In the RDF format, the values are represented as two forms - simple value and full value. Simple value is always a literal or IRI, and is used as direct value that is convenient to search, index and match. The full value contains additional information about the value, such as ranges, precision, calendar used, etc. Note that while for many queries simple values will be enough, for other, more complex values, only full values will be adequate.

If the statement has a value (i.e. is not set to novalue ) then the simple value will always be present.

Full values are represented as nodes having prefix wdv: and the local name being the hash of the value contents (e.g. wdv:382603eaa501e15688076291fc47ae54 ). There is no guarantee of the value of the hash except for the fact that different values will be represented by different hashes, and same value mentioned in different places will have the same hash. Value node has type wikibase:Value . The content of the node is defined by the type of the value (see below).

Example of the value node:

The following describes the handling of each kind of value, depending on the type of the value and the type of the property. Note that not all aspects of the data model are represented in RDF currently, some aspects that are currently unused (such as units or before/after values for dates) are omitted since they currently do not carry any useful information. This may change in the future if/when these aspects come into use by Wikidata.

Strings have value type string and property type string .

String is represented as a string literal. Strings only have simple value.

Media on commons: have value type string and property type commonsMedia .

Commons media is represented as an IRI with the full Commons resource URL, derived from the Commons filename in the underlying data item. E.g.: < http://commons.wikimedia.org/wiki/Special:FilePath/Universe Photo.svg > . It has only simple value.

URL values have value type string and property type url .

URL is represented as a an IRI matching the URL string (e.g. < http://www.wikidata.org/ >) . It has only simple value.

External Id values have value type string and property type external-id . They are represented by a string literal. It has only simple value.

If the property has URL formatter for RDF configured, the RDF will also have normalized value, e.g.:

Wikibase Entity Id values have value type wikibase-entityid and property type wikibase-item .

The entity is represented by its IRI, e.g. wd:Q2 . It has only simple value.

Monolingual text values have value type monolingualtext and property type monolingualtext .

The text is represented as a string literal with language tag. It has only simple value.

Coordinate text values have value type globecoordinate and property type globe-coordinate .

The simple value of the coordinate is the WKT string with the coordinates, with type geo:wktLiteral , e.g.: "Point(35.3 12.93)"^^ geo:wktLiteral . The order of the coordinates in WKT is longitude, latitude (since format version 0.0.2).

The full value has latitude, longitude and precision as double, and the globe as IRI.

Example:

Quantity values have value type quantity and property type quantity .

The simple value of the quantity is the specified amount, as a decimal literal.

The full value includes amount, unit URI (the default for unit-less values being http://www.wikidata.org/entity/Q199 ), and optionally upper and lower bound. If no upper an lower bound are given, the uncertainty of the quantity is undefined. Exact values are represented by quantities that have the same value for amount, upper bound and lower bound.

Example:

Time values have value type time and property type time .

The simple value of the time value is either datetime value of type xsd:dateTime , if the value can be converted to Gregorian date in ISO format, or a string as represented in the database, if not. The xsd:dateTime dates follow XSD 1.1 standard , which uses the proleptic Gregorian calendar , and represents the year 1 BCE as +0000. This is in contrast the JSON representation of Julian and Gregorian dates , which follows the traditional year numbering, representing the year 1 BCE as -0001.

The full value includes the simple value above under wikibase:timeValue , precision and timezone as integers and calendar model as IRI. Note that the calendar model is the original values calendar model even if wikibase:timeValue was converted to Gregorian.

Example:

Some values can be represented in several forms, depending on the purpose. For example, length can be expressed in different units - feet, inches, meters, miles, etc. In order to provide means to unify these forms and thus make data more friendly for automatic processing, the normalized values are introduced, which represent diverse data in a unified way.

Right now, the only value normalization that is supported is converting units for quantities into base units - e.g. length to meters. In the future, more units and more normalizations may be added, which will be documented here. The conversion table is available on the Mediawiki gerrit if needed.

The only normalized simple values are external IDs (see below).

Normalized quantity values are value nodes that are parallel to the original data nodes but represented in base units. They are connected to their parent nodes by predicates with prefix having "v" replaced with "n" - i.e. psn: , prn: and pqn: , for example:

Original quantity value is connected to the normalized value by wikibase:quantityNormalized predicate:

The normalized value has wikibase:quantityNormalized pointing to itself.

If the value is already normalized - i.e. is expressed in base units - then both "v" and "n" predicates point to the same value, and wikibase:quantityNormalized for this value points to itself.

Quantities with no units or with units that are not normalizable (have no base unit they can be reduced to) do not have normalized predicates and normalized values and do not include wikibase:quantityNormalized .

The recommendation is to have no more than one base unit per property. Base units depend on Wikibase configuration and usually are chosen to represent universally accepted standard units, such as SI units.

For external IDs, normalization converts string value to URL, if the URL formatter for that purpose is defined in property data (via canonicalUriProperty setting), then the normalized value will be listed as wdtn: value for truthy values, and as normalized value for the statements in psn: , prn: and pqn: predicates, depending on the context where the value appears.

Wikibase data model has two special kinds of snaks - PropertySomeValueSnak, specifying a value that exists but whose identity or value is unknown, and PropertyNoValueSnak, specifying that a value does not exist.

Unknown value is represented as RDF blank node in both simplified and full statements:

Novalue is represented not by a regular value but as a class of the entity or statement or reference, with prefix wdno: and the name of the property. Example:

The entity has a wdno: class if it has a truthy novalue statement for that property. Novalue in the main snak or qualifiers of a statement corresponds to a wdno: class on the statement node, and novalue in a reference snak corresponds to a wdno: class on the reference node.

The classes for wdno: are defined as follows:

The links are represented as set of predicates describing the link URL. The type of the node is schema:Article and it linked with the entity via schema:about predicate.

Badges are described with wikibase:badge  predicates. schema:name predicate holds the plain-text name of the article, in the language of the linked wiki.

Example:

The subject URL is composed from the language site prefix and the article name, URL-encoded according to RFC 3986 , e.g.:

More specifically, encoding used is as follows:

Redirected entities are implemented as owl:sameAs predicates, for example if Q6 redirects to Q1, the dump would be:

The prefixes are used in RDF formats that allow short prefixes (such as Turtle and RDF). For other formats, the full URL is used.

All prefix URLs that do not contain hostname are prefixed with the hostname of the generating wiki. All prefix URLs that contain hostname are fixed and do not depend on generating wiki.

wd

:

Q2

a

wikibase

:

Item

wdata:

Q2



schema:

about



wd:

Q2



.

wd:

Q2



p:

P9



wds:

Q2-82a6e009-4f93-28dc-3555-38bbfc3afe6a

wds

:

Q2-a4078553-4ec1-a64a-79e7-c5b5e17b2782

  
a

wikibase

:

Statement

wdv

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

a

wikibase

:

Value

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

prov

:

wasDerivedFrom

wdref

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

.

wdref

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

a

wikibase

:

Reference

.

wd:

Q2

  

wdt:

P9



<http://acme.com/>

wd

:

Q2
  
wdtn

:

P9

<http://acme.com/ABCDE>

wd:

Q2



p:

P9



wds:

Q2-82a6e009-4f93-28dc-3555-38bbfc3afe6awd

wd

:

Q2

a

wdno

:

P9

.

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

ps

:

P8

"-13000000000-01-01T00:00:00Z"

^^

xsd

:

dateTime

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

psv

:

P8

wdv

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

psn

:

P8

wdv

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

pq

:

P8

"-13000000000-01-01T00:00:00Z"

^^

xsd

:

dateTime
 

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

pqv

:

P8

wdv

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

wds

:

Q3-24bf3704-4c5d-083a-9b59-1881f82b6b37

pqn

:

P8

wdv

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

wdref

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

pr

:

P8

"-13000000000-01-01T00:00:00Z"

^^

xsd

:

dateTime

wdref

:

87

d

0

dc

1

c

7847

f

19

ac

0

f

19

b

e978015

dfb

202

cf

59

a



prv:

P8



wdv

:

87

d

0

dc

1

c

7847

f

19

ac

0

f

19

b

e978015

dfb

202

cf

59

a

wdref

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

prn

:

P8

wdv

:

87d0dc1c7847f19ac0f19be978015dfb202cf59a

Standard prefixes used:

This list can be used for queries in SPARQL:

This compiles the list of all objects and predicates that are internal to the format. For the meaning of the prefixes, see the prefixes list.

Italicized names mean that any property name can be substituted instead of example name P123.

The following predicates are used in deep values for the values of specific types. All these predicates have the domain of wikibase:Value and the range depending on type below.

The Wikidata query service has the data in the format described above, but there are small differences that can be important while writing SPARQL queries:

See also SPARQL query examples for how to query the data using WDQS service.