Provenance is information about entities, activities, and people involved in producing a piece of data or thing, which can be used to form assessments about the data or thing's quality, reliability or trustworthiness. PROV-DM is the conceptual data model that forms a basis for the W3C provenance (PROV) family of specifications. This document specifies PROV-JSONLD, a serialization of PROV in JSON, which exploits JSON-LD to define a semantic mapping so it can also be processed as Linked Data. Overall, PROV-JSONLD is designed to be suitable for interchanging provenance in Web and Linked Data applications, to offer a natural encoding of provenance for its targeted audience, and to allow for fast processing.
OWL Ontology Extension to support PROV-JSONLD provext.ttl
Status of This Document
This section describes the status of this
document at the time of its publication. A list of current W3C
publications can be found
in the W3C technical reports index at
https://www.w3.org/TR/.
A member submission by King's College London
By publishing this document, W3C acknowledges that the Submitting Members have made a formal Submission request to W3C for discussion. Publication of this document by W3C indicates no endorsement of its content by W3C, nor that W3C has, is, or will be allocating any resources to the issues addressed by it. This document is not the product of a chartered W3C group, but is published as potential input to the W3C Process. A W3C Team Comment has been published in conjunction with this Member Submission. Publication of acknowledged Member Submissions at the W3C site is one of the benefits of W3C Membership. Please consult the requirements associated with Member Submissions of section 3.3 of the W3C Patent Policy. Please consult the complete list of acknowledged W3C Member Submissions.
1. Conformance
As well as sections marked as non-normative, all authoring guidelines, diagrams, examples, and notes in this specification are non-normative. Everything else in this specification is normative.
The key words MAY, MUST, RECOMMENDED, REQUIRED, and SHOULD in this document
are to be interpreted as described in
BCP 14
[RFC2119] [RFC8174]
when, and only when, they appear in all capitals, as shown here.
2. Introduction
Since their release in 2013, the PROV Recommendations [PROV-OVERVIEW] by the World Wide Web Consortium (W3C) have been adopted by flagship deployments such as the Global Change Information System, the Gazette in the UK, and other Linked Data sets. PROV, which is used as the data model to describe the provenance of data, is made available in several different representations: PROV-N [PROV-N], PROV-XML [PROV-XML], or in an RDF serialization using the PROV Ontology [PROV-O]. The latter is most suitable for Linked Data [LINKED-DATA], given that it can readily be consumed by existing Semantic Web tools and comes with the semantic grounding provided by PROV-O [PROV-O].
Subsequently, the PROV-JSON [PROV-JSON] serialization has gained traction, despite simply being a member submission, and not having gone through the various stages of a standardization activity. We conjecture that the primary reason for this is that many web applications are built to be light-weight, working mainly with simple data formats such as JSON [RFC8259].
The very existence of all these serializations is a testament to the approach to standardization taken by the Provenance Working Group, by which a conceptual data model for PROV was defined, the PROV data model [PROV-DM], alongside its mapping to different technologies, to suit users and developers. However, the family of PROV specifications lacks a serialization capable of simultaneously addressing all of the following requirements.
[Lightweight] A serialization MUST support lightweight Web applications.
[Natural] A serialization MUST look natural to its targeted community of users.
[Semantic] A serialization MUST allow for semantic markup and integration with linked data applications.
[Efficient] A serialization MUST be efficiently processable.
In our view, none of the existing PROV serializations supports all these requirements simultaneously. While PROV-JSON is the only serialization to support lightweight web applications, it does not have any semantic markup, its internal structure does not exhibit the natural structure of the PROV data structures, and its grouping of expressions per categories (e.g. all entities, all activities, ...) is not conducive to incremental processing. The RDF serialization compatible with PROV-O has been architected to be natural to the Semantic Web community: all influence relations have been given the same directionality, consistently aligned with their time ordering, but the decomposition of data structures (essentially n-ary relations) into individual triples, which can occur anywhere in the serialization, is not conducive to efficient parsing. It is reasonable to say that the world has moved on from XML, while the PROV-N notation was aimed at humans rather than efficient processing.
JSON-LD [JSON-LD] allows a semantic structure to be overlaid over a JSON structure [RFC8259], thereby enabling the conversion of JSON serializations into linked data. This was exploited in an early version of this work [IPAW-POSTER], which applied the JSON-LD approach to a JSON serialization of PROV. The solution did not lead to a natural encoding of the PROV data structure because a property occurring in different types of JSON objects had to be named differently so that it could be mapped to the appropriate RDF property; we see here that what is natural in JSON is not necessarily natural in RDF, and vice-versa. The ability to define contextual mappings was introduced in JSON-LD 1.1 [JSON-LD11] and is a key enabler of this specification, allowing for the same natural PROV property names to be used in different contexts while still maintaining their correct mappings to the appropriate RDF properties.
Thus, this specification proposes PROV-JSONLD, a PROV serialization compatible with [PROV-DM] that addresses all of our 4 key requirements. It is, first and foremost, a JSON structure supporting lightweight Web applications. It is structured so that each PROV expression is encoded as a self-contained JSON object and, therefore, is natural to JavaScript programmers. Exploiting JSON-LD 1.1, we defined contextual semantic mappings, allowing PROV-JSONLD to be seen as linked data. And finally, PROV-JSONLD allows for efficient processing since each JSON object can be readily mapped to a data structure without requiring unbounded lookaheads or search within the data structure.
In the rest of this document, we illustrate PROV-JSONLD, we characterize its structure using a JSON Schema [JSON-SCHEMA], we define its semantic mappings using JSON-LD 1.1, and we outline the interoperability testing we put in place to check its compatibility with the PROV data model.
2.1 Namespace
The following namespaces prefixes are used throughout this document.
Table 1: Prefix and Namespaces used in this specification
All other namespace prefixes are used in examples only. In particular, IRIs starting with "http://example.com" represent some application-dependent IRI [RFC3987]
3. Example
This section is non-normative.
We assume the reader to be familiar with PROV, JSON, and JSON-LD.
To illustrate the PROV-JSONLD serialization, we consider a subset of the example of [PROV-PRIMER], depicted below. It can be paraphrased as follows: agent Derek was responsible for composing an article based on an existing dataset.
Figure 1: Provenance expressing that Derek was responsible for composing an article based on a data set.
The PROV-JSONLD representation of this example can be seen in Example 1. At the top level, a PROV-JSONLD document is a JSON object with two properties
@context
and
@graph, as per JSON-LD. A context contains mappings of prefixes to namespaces, and also an explicit reference to https://openprovenance.org/prov-jsonld/context.jsonld — the JSON-LD 1.1 context defining the semantic mapping for PROV-JSONLD. (This context is fully described in section 5.)
The @graph property has an array of PROV expressions as value. Each PROV expression is itself a JSON object with at least a @type property (for instance, Entity, Agent or Derivation). Each of these PROV expressions provides a description of a resource. Some of these resources have an identity provided by the @id property (for instance, ex:article1 or ex:derek). Other resources are anonymous and, therefore, do not have a property @id, for instance, the Derivation between the dataset and the article.
PROV expressions can be enriched with various properties. Some properties are predefined by PROV-JSONLD such as activity and agent in a Association. Further PROV attributes are allowed, for instance type with an array of further types, to better describe the resource. Others may be defined in a different namespace such as foaf:givenName, for which we expect the prefix foaf to be declared in the @context property.
The property @type is mandatory and is associated with a single value, expected to be one of the predefined PROV expressions. From an efficiency viewpoint, this property is critical in determining which internal data structure a PROV expression should map to, and therefore, facilitates efficient processing. On the contrary, type is optional and can contain as many types as required; their order is not significant.
The production rules for qualified names are more complex than the simple regular expression outlined here. A post-processor will need to check that qualified names comply with the definition in [PROV-N].
Typed values (typed_value) are JSON objects with properties @value and @type. String values (lang_string) are JSON objects with properties @value and @language.
With these preliminary definitions in place, we can now present the specification of PROV-JSONLD's core data structures.
4.2 prov:Entity
In the Schema for prov:Entity, an entity MUST contain an identifier (property @id) and a property @type with value Entity. It MAY contain further type information (property type, see PROV-DM prov:type), a location (property location, see PROV-DM prov:location), a label (property label, see PROV-DM prov:label), or other properties with an explicit prefix (see PROV-DM entity attributes). (The presence of a colon ":" in the patternProperties element forces all other properties to have the structure of a prefix, a colon, and a local name.)
In the Schema for prov:End, an end MUST contain a property @type with value End.
It SHOULD contain an activity that was ended (property activity, see PROV-DM activity); it MAY contain an ender activity (property ender, see PROV-DM ender) and a triggering entity (property trigger, see PROV-DM trigger).
It MAY also contain an identifier (property @id), a time (property time, see PROV-DM time), a location (property location, see PROV-DM prov:location), a role (property role, see PROV-DM prov:role), further type information (property type, see PROV-DM prov:type), a label (property label, see PROV-DM prov:label), or other properties with an explicit prefix (see PROV-DM end attributes).
In the Schema for prov:Specialization, a specialization MUST contain a property @type with value Specialization.
It SHOULD contain a specific entity (property specificEntity, see PROV-DM specificEntity) and a general entity (property generalEntity, see PROV-DM generalEntity).
It MAY contain an identifier (property @id), further type information (property type, see PROV-DM prov:type), a label (property label, see PROV-DM prov:label), or other properties with an explicit prefix. (There is no equivalent for those properties in PROV-DM, see section Interoperability.)
In the Schema for prov:Alternate, an alternate MUST contain a property @type with value Alternate.
It SHOULD contain a first alternate (property alternate1, see PROV-DM alternate1) and a second alternate (property alternate2, see PROV-DM alternate2).
It MAY contain an identifier (property @id), further type information (property type, see PROV-DM prov:type), a label (property label, see PROV-DM prov:label), or other properties with an explicit prefix. (There is no equivalent for those properties in PROV-DM, see section Interoperability.)
In the Schema for prov:Membership, a membership MUST contain a property @type with value Membership.
It SHOULD contain a collection (property collection, see PROV-DM collection) and a single entity or an array of them (property entity, see PROV-DM entity).
It MAY contain an identifier (property @id), further type information (property type, see PROV-DM prov:type), a label (property label, see PROV-DM prov:label), or other properties with an explicit prefix. (There is no equivalent for those properties in PROV-DM, see section Interoperability.)
In the Schemas for prov:Bundle and prov:Document, a bundle and a document MUST contain a property @type with value Bundle and Document respectively, a context @context, and set of PROV expressions @graph. The names of the properties @context and @graph are specified by JSON-LD [JSON-LD11]. In addition, a bundle must contain an identifier (property @id).
Bundles contain statements (definition prov:Statement), whereas documents contain statements or bundles (definition prov:StatementOrBundle); see their specification in Schemas for prov:Statement and prov:StatementOrBundle.
Schemas for prov:Statement and prov:StatementOrBundle
The type Context is described in the next section.
4.20 Context
Finally, contexts are defined according to Schema for Context. They take the shape of an array, containing either mappings of prefixes to IRIs or IRIs to further JSON-LD contexts.
In this section, we provide a description of the JSON-LD context to map the PROV-JSON structures to linked data. Full details of the context can be found in Appendix B.
5.1 Introduction: Qualification Pattern
The Ontology PROV-O [PROV-O] defines the Qualification Pattern, which restates a binary property between two resources (referred to as an unqualified influence relation) by using an intermediate class that represents the influence between two resources. This new instance, in turn, can be annotated with additional descriptions of the influence that one resource had upon another. The following figure, borrowed from [PROV-O], summarises the PROV relations, and how they are encoded in RDF using the Qualification Pattern.
Note that the figure does not include the Qualification Pattern for Influence; in addition, PROV-O does not define the Qualification Pattern for specialization, alternate and membership.
Figure 2 (taken from [PROV-O]):
Illustration of the properties and classes to use (in blue)
to qualify the binary influence relations (dotted black).
The diagram depict entities as ovals, activities as rectangles, and agents as pentagons. The Qualified Resource is represented as a left-pointy shape: in PROV-JSONLD, a Qualified Resource is represented as a JSON Object.
5.2 Default and generic Context Elements
The following JSON properties have a default meaning, unless they are redefined in a specific context of a PROV-JSONLD document:
entity,
activity and
agent
respectively map to PROV-O object properties
prov:entity,
prov:activity and
prov:agent.
The following JSON properties have the same meaning in all contexts of a PROV-JSONLD document:
role,
type,
label and
location
respectively map to the RDF properties
prov:hadRole,
rdf:type,
rdfs:label, and
prov:atLocation.
In the mapping Context for Activity, the JSON properties startTime and endTime map to the RDF data properties prov:startedAtTime and prov:endedAtType, respectively, and have a value coerced to datatype xsd:dateTime.
The mapping Context for Derivation supports the Qualification Pattern of Figure 2, g. Each of the JSON properties
generatedEntity,
usedEntity,
activity,
generation, and
usage
maps to an object property: namely,
prov:qualifiedDerivation,
prov:entity,
prov:hadActivity,
prov:hadGeneration, and
prov:hadUsage, respectively.
The mapping Context for Attribution supports the Qualification Pattern of Figure 2, i. The JSON property
entity
maps to the object property
prov:qualifiedAttribution.
The mapping Context for Association supports the Qualification Pattern of Figure 2, j. The JSON properties
activity and
plan
map to the object properties
prov:qualifiedAssociation and
prov:hadPlan, respectively.
The mapping Context for Delegation supports the Qualification Pattern of Figure 2, h. The JSON properties
responsible,
delegate and
activity
map to the object properties
prov:agent,
prov:qualifiedDelegation and
prov:hadActivity, respectively.
The mapping Context for Usage supports the Qualification Pattern of Figure 2, a.
The JSON properties
activity and
time
map to the object property
prov:qualifiedUsage and the data property
prov:atTime, respectively. The value of the latter is coerced to datatype
xsd:dateTime.
The mapping Context for Generation supports the Qualification Pattern of Figure 2, b.
The JSON properties
entity and
time
map to the object property
prov:qualifiedGeneration and the data property
prov:atTime, respectively. The value of the latter is coerced to datatype xsd:dateTime.
The mapping Context for Invalidation supports the Qualification Pattern of Figure 2, c.
The JSON properties
entity and
time
map to the object property
prov:qualifiedInvalidation and the data property
prov:atTime, respectively. The value of the latter is coerced to datatype xsd:dateTime.
The mapping Context for Start supports the Qualification Pattern of Figure 2, e.
The JSON properties
activity,
trigger,
starter, and
time
map to the object properties
prov:qualifiedStart,
prov:entity,
prov:hadActivity, and the data property
prov:atTime, respectively. The value of the latter is coerced to datatype
xsd:dateTime.
The mapping Context for End supports the Qualification Pattern of Figure 2, f.
The JSON properties
activity,
trigger,
ender, and
time
map to the object properties
prov:qualifiedEnd,
prov:entity,
prov:hadActivity, and the data property
prov:atTime, respectively. The value of the latter is coerced to datatype
xsd:dateTime.
The mapping Context for Communication supports the Qualification Pattern of Figure 2, d.
The JSON properties
informed and
informant
map to the object properties
prov:qualifiedCommunication and
prov:activity, respectively.
In the mapping Context for Influence, the JSON properties
influencee and
influencer
map to the object properties
prov:qualifiedInfluence and
prov:influencer, respectively.
While [PROV-O] does not define a Qualification Pattern for Specialization, for uniformity and usability reasons, we adopt a similar mapping as other PROV relations, via a Qualification Pattern (see Definition Context for Specialization). However, the mapping is to new classes and properties in the PROV extension namespace (denoted by the prefix provext). See Section 6 for interoperability considerations.
The JSON properties
specificEntity and
generalEntity
map to the object properties
provext:qualifiedSpecialization and
provext:generalEntity, respectively.
While [PROV-O] does not define a Qualification Pattern for Alternate, for uniformity and usability reasons, we adopt a similar mapping as other PROV relations, via a Qualification Pattern (see Definition Context for Alternate). However, the mapping is to new classes and properties in the PROV extension namespace (denoted by the prefix provext). See Section 6 for interoperability considerations.
The JSON properties
alternate1 and
alternate2
map to the object properties
provext:qualifiedAlternate and
provext:alternate, respectively.
While [PROV-O] does not define a Qualification Pattern for Membership, for uniformity and usability reasons, we adopt a similar mapping as other PROV relations, via a Qualification Pattern (see Definition Context for Membership). However, the mapping is to new classes and properties in the PROV extension namespace (denoted by the prefix provext). See Section 6 for interoperability considerations.
The JSON properties
collection and
entity
map to the object properties
provext:qualifiedMembership and
provext:member, respectively.
There are differences between PROV-DM and PROV-O in terms of the level of requirements set on some expressions. For instance, PROV-DM mandates the presence of an entity in a generation, whereas it defines an activity as optional. Compliance requirements are not the same in PROV-O as one could define a qualified generation with an activity but without an entity. Experience shows that there may be good reasons why a generation may not refer to an entity; for instance, because the recorded provenance is not "complete" yet, and further provenance expressions still need to be asserted, received or merged; in the meantime, we still want to be able to process such provenance, even if "incomplete". Thus, in PROV-JSONLD, the presence of an entity and an activity in a generation expression is RECOMMENDED (we use the term SHOULD), while other properties are optional (we use the term MAY), and its @type is REQUIRED (we use the term MUST).
IC2:
In PROV-DM, all relations are n-ary except for specialization, alternate and membership, which are binary, meaning that no identifier or extra properties are allowed for these. In PROV-O, this design decision translates to the lack of qualified relations for specialization, alternate and membership. In PROV-JSONLD, in order to keep the regular structure of JSON objects and the natural encoding of relations, but also to ensure the simplicity and efficiency of parsers, these three relations are encoded using the same pattern as for other relations. Therefore, their mapping to RDF via the JSON-LD context relies on a PROV extension namespace (denoted by the prefix provext) in which classes for Specialization, Alternate and Membership are defined. The PROV-JSONLD serialization also allows for identifier and properties to be encoded for these relations.
IC3:
The notion of a PROV document is not present in PROV-DM or PROV-O, but is introduced in PROV-N as a housekeeping construct, and is defined in PROV-XML as the root of a PROV-XML document. A document in PROV-JSONLD is also a JSON object, allowing for a JSON-LD @context property to be specified.
IC4:
The PROV-JSONLD specification does not introduce constructs for some PROV subtypes and subrelations, such as prov:Person, prov:Organization, prov:SoftwareAgent, prov:Collection, or prov:Quotation, prov:PrimarySource, prov:Revision. Instead, the example of Section 3 illustrates how they can be accommodated within the existing structures. We copy below an agent expression of type prov:Person and a derivation of type prov:Revision. These subtypes and subrelations are specified inside the prov:type property. PROV-XML offers a similar way of encoding such subtypes and subrelations, alongside specialized structures. We opted for this single approach to ensure simplicity and efficiency of parsers.
The interoperability of the PROV-JSONLD serialization can be tested in different ways:
In a roundtrip testing, consisting of the serialization of an internal representation in some programming language to PROV-JSONLD, followed by deserialization from PROV-JSONLD back to the same programming language, the source and target representations are expected to be equal.
Likewise, in a roundtrip testing, consisting of the serialization of an internal representation in some programming language to PROV-JSONLD, followed by a conversion of PROV-JSONLD to another RDF representation such as Turtle, followed by a reading of the Turtle representation back to the same programming language, the source and target representations are also expected to be equal.
Both interoperability tests have been implemented in the Java-based ProvToolbox, with:
C. PROVEXT: PROV Ontology Extension for PROV-JSONLD
We provide here a minimal definition of the classes and properties introduced in provext in the context of PROV-JSONLD. They allow the qualification pattern to be applied to Membership, Specialization and Alternate. For each, we define one class and two object properties.
