Wikidata:Data model

Wikidata represents entities as data items (e.g. Tim Berners-Lee (Q80) and CERN (Q42944) are data items). Knowledge about data items is represented via statements, whose basic structure consists of a subject, a predicate and an object. For example, Tim Berners-Lee (Q80)employer (P108)CERN (Q42944).

  • The subject of a statement is usually a data item — in this case, Tim Berners-Lee (Q80).
  • The predicate of a statement is always a property — in this case, employer (P108).
  • The object of a statement is a value of the data type of the property — in this case, an item, CERN (Q42944).

The property used in a statement determines both the meaning of the statement (i.e. the nature of the relationship between the subject and the object), as well as which values may be used, as specified by its data type.

For example, in the example above we used the property employer (P108), whose values must have the data type Item, allowing a data item to be set as the object of the statement (in the case of our example, CERN (Q42944)).

An example of a property with a different data type is start time (P580), whose values must be of data type Point in time, so it can only be used to state a point in time.

Wikidata also allows statements to be qualified with further properties, which are called qualifiers. For example, we might state Tim Berners-Lee (Q80)employer (P108)CERN (Q42944)start time (P580)June 1980end time (P582)December 1980.

The information on this page is not required to contribute to Wikidata or to consume Wikidata. To learn about contributing/consuming Wikidata, please refer to the pages Wikidata:Introduction and Wikidata:Data access respectively.

Three levels of data models

edit

Wikidata is powered by the Wikibase software. While Wikibase defines 12 data types by default, it does not come with any property out of the box. Wikidata, however, has 12,248 properties, which have all been created specifically for Wikidata and are defined within Wikidata itself. (Don't worry about that large number, 75% of these properties are just external identifiers, i.e. links to items in other databases.)

When we speak of a "data model" in the context of Wikidata, it can actually refer to one of three things:

All of these different data models are described on different pages:

Note that Wikidata has no central authority that decides how data should be modeled, instead that question is decided collaboratively by the community through public discussion. The data model of Wikidata has evolved over time and is very much still evolving: new data types can be introduced, new properties are being proposed and created, problematic properties get deprecated and there is an ongoing effort to better describe how properties are meant to be used via property constraints and entity schemas.


Data model of Wikibase

edit

The data model of Wikidata is based on the data model of Wikibase, which is described very technically in the specification and more accessibly in the primer to the Wikibase data model.

Wikidata extends the Wikibase data model via extensions. Most notably WikibaseLexeme adds three entity types for lexicographical data (Lexeme, Form and Sense), as described in the WikibaseLexeme data model. Wikidata uses several extensions to add more data types to Wikibase, as described in Wikidata:Data model#Data types.

Data types

edit

The data types of Wikidata are described at Help:Data type and listed at Special:ListDatatypes. Wikidata extends the data types of Wikibase via the following three extensions:

This is possible because the data types of Wikibase are extensible. The introduction of more data types can be proposed on Phabricator.

The Wikibase data model has a canonical representation in JSON, which is further described at Wikidata:JSON format.

Note that several data types have limitations, which are listed at Help:Data type.

Also note that there is no clear semantical difference between String and External identifier ... several string properties are external identifiers and formatter URL (P1630) works for both.

Ranks

edit

Every statement in Wikibase has one of three ranks (normal, deprecated or preferred). For the semantics of these ranks please refer to Help:Ranking#Usage.

No value and unknown value

edit
  • SPno value means that no such value exists (≡ ¬∃ X (SPX))
  • SPunknown value can mean any of the following:
    • the value was once known but has been lost to time (e.g. Paolo Baronni (Q7132144)date of birth (P569)unknown value)
    • the exact value has never been known and might not ever be known (e.g. star (Q523)quantity (P1114)unknown value)
    • the Wikidata contributor who made the statement knows the value exists but doesn't know it personally
    • the value is a known object, but there's no Wikidata item about the object (perhaps because it's not notable).


Order of values

edit

While Wikibase always stores values in a specific order (insertion order by default), the order of values generally does not imply any semantics. Semantic order is instead expressed via qualifiers, for example:

Note that the order expressed via qualifiers does not necessarily match the order of values in the user interface or the API because these interfaces simply return values in the serialization order, which may or may not match the semantic order expressed by the qualifiers.[2]

Fundamental entities

edit

The fundamental properties of Wikidata are described in

  1. Fundamental properties.

For more information and people interested in the ontology of Wikidata, please refer to the Ontology WikiProject.

Fundamental properties

edit

Note: This section assumes that you are familiar with logical operators, for a less technical explanation please refer to Help:Basic membership properties. The three arguably most important properties of Wikidata are based on RDF Schema, which is described in the RDF Schema specification.

These properties have the following semantics:

Please note that subclass of (P279) and subproperty of (P1647) are both transitive properties:

Another important property is inverse property (P1696), which is equivalent to owl:inverseOf and carries the following semantics:

Restrictiveness of qualifiers

edit

Qualifiers</tvar> can be either restrictive or non-restrictive. Restrictive qualifiers change the meaning or scope of a statement, they have to be taken into account by data consumers that want to correctly interpret Wikidata statements. Non-restrictive qualifiers on the other hand just add additional information that can be safely disregarded without changing the meaning or scope of the statement.

Examples for restrictive qualifiers are:

The restrictiveness of properties when used as a qualifier is currently modeled via instance of (P31)restrictive qualifier (Q61719275) and instance of (P31)non-restrictive qualifier (Q61719274) (note that you as always have to take the transitivity of instance of (P31) into account).

Unfortunately some properties aren't clear-cut and can be both restrictive as well as non-restrictive when used as a qualifier, so we can group qualifier properties into four categories:

Negation

edit

Wikibase does not have built-in support for negation, negation therefore has to be modeled with separate properties. For example has part(s) (P527) can be negated with does not have part (P3113). Such negating properties only exist for a few properties. When the need for a new negating property arises, it may be proposed.

The semantics of negating properties are modeled via negates property (P11317), as follows:

Whether or not a property expresses the absence of something is currently modeled via instance of (P31)Wikidata property to express the absence of something (Q115449020).

Differences from OOP

edit

Contrary to object-oriented programming there is nothing preventing an entity from being both an instance as well as a class.

Furthermore an entity can be an instance of multiple classes, as well as a subclass of multiple classes.

Lastly you might expect that an instance automatically inherits all statements from its parent classes, however that is explicitly not the case, as explained in Wikidata:Data model#Inheritance.

Inferring classes

edit

Properties may specify class of non-item property value (P10726) which has the semantics:

Classes can be defined to be a union or a disjoint union of other classes with union of (P2737) and disjoint union of (P2738) respectively. Their concrete semantics are as follows:

Let's define  .

Classes may specify union of (P2737) which has the semantics:

Classes may specify disjoint union of (P2738) which has the semantics:

Inheritance

edit

If you are familiar with object-oriented programming, you might expect that instances of a class inherit the statements of a class. This is generally not the case. For example just because horse (Q726)studied in (P2579)hippology (Q1157006) and Apology (Q4780432)instance of (P31)horse (Q726) does not mean that Apology (Q4780432)studied in (P2579)hippology (Q1157006). However there are some properties that are likely to be inherited:

Property Inverse property
has part(s) (P527) part of (P361)
has characteristic (P1552) none
has cause (P828) has effect (P1542)
uses (P2283) used by (P1535)

For example public website (Q115449506)part of (P361)World Wide Web (Q466) and YouTube (Q866)instance of (P31)public website (Q115449506) can be used to correctly infer YouTube (Q866)part of (P361)World Wide Web (Q466).

When attempting to make such inferences don't forget to take ranks, restrictive qualifiers and negation into account, as explained in Wikidata:Data model#Does a statement apply?.

Does a statement apply?

edit

The following is an attempt at outlining a strategy to decide whether a particular statement applies to a given entity:

  1. Statements ranked as deprecated have been superseded and therefore no longer apply.
  2. Statements with a restrictive qualifier only apply with regards to the respective qualifier.
  3. Statements of certain properties are likely to be inherited (see inheritance). Note however that instances or intermediary classes may negate statements inherited from a parent class, as described in negation.


Reflexive statements

edit

APA has unclear semantics if A is a class, it could mean:

  1. an instance of A has a relation P to another instance of A (which may or may not be the same instance)
  2. an instance of A has a relation P to a different instance of A (which cannot be the same instance)
  3. an instance of A has a relation P to itself

See object is for a proposal to introduce a qualifier property to differentiate these cases.

Format string properties

edit

Wikidata has several format string properties, such as formatter URL (P1630), DOI formatter (P8404) and URN formatter (P7470).

The formatting mechanism of these properties and what kind of values they produce is currently not stated in a machine-readable manner, however that might change with the introduction of the proposed format string properties.


Property constraints

edit

Wikidata employs property constraints to combat property misuse. Property constraints are implemented by Extension:WikibaseQualityConstraints and are stated on properties via property constraint (P2302) since 2017. [3] The violation of such property constraints is directly displayed in the Wikidata user interface.

More complex property constraints can be implemented as SPARQL queries and placed with {{Complex constraint}} on property talk pages. The violation of such complex constraints is periodically reported by a bot on pages within the Category:Complex constraint violation reports category.

For more information about property constraints, please refer to the help portal and the property constraints WikiProject.

Topic-specific data models

edit

Wikidata covers many topics, such as art, biology, countries, cities, monuments, movies, people, software, websites, writings, etc. All entities of these topics that are notable somehow need to be represented as data items with statements. So which statements should be made for a specific entity type and which properties should be used for these statements? The answers to these questions are subject to the topic-specific data model that should be used for the specific topic. So, which data model should be used for a given topic? That is decided collaboratively by the Wikidata community through public discussion. The discussions and efforts about a specific topic in Wikidata are organized via WikiProjects.

Where can you find topic-specific data models?

Entity schemas

edit

An alternative approach to property constraints is using the Shape Expressions data modelling language. For Wikidata such schemas can be stored within the EntitySchema:* namespace on the wikidata.org wiki (which is enabled by the EntitySchema MediaWiki extension). Note that the effort to establish such schemas for Wikidata is very much ongoing: the Shape Expression for class property proposal is currently on hold because the EntitySchema data type is not yet implemented.[5]

For more information about Wikidata Schemas, please refer to the Schemas WikiProject.


See also

edit

References

edit
  1. Items can have labels, descriptions, aliases and sitelinks, statements have a rank and can have qualifiers and references, and values can also be specified as no value or unknown value.
  2. Phabricator task T173432: Sort claims of a property in meaningful way
  3. Phabricator task T102759: Migrate constraints from property talk pages to statements on properties
  4. it is possible that such variety will be standardized in the future
  5. Phabricator task T214884: linking Schemas in statements