Generate SHACL shapes from data¤

This workflow task generates SHACL (Shapes Constraint Language) node and property shapes by analyzing the instance data of a knowledge graph. The generated shapes describe the classes the data uses and the properties each class is used with.

The data graph and the shape catalog are both chosen by IRI parameter rather than by workflow connection, so the task has neither an input nor an output port. It is a terminal step: it reads and writes graphs directly and hands nothing on to a following task.

What is generated¤

• A node shape for every class that instances are typed with.
• A property shape for every property those instances use: object properties in the subject → object direction, object properties in the object ← subject direction, whose names carry a ← prefix, and datatype properties, whose values are literals.

Every shape gets an IRI derived from a UUID5 of the class or property it describes, so within one catalog the same class or property always maps to the same shape. Names and labels come from the title eccenca Corporate Memory resolves for that class or property. A description follows wherever one can be resolved - the rdfs:comment, dcterms:description or skos:definition of the class or property, looked up wherever it is defined, so a vocabulary graph counts as well as the data graph. A class or property that nothing describes gets no sh:description, and neither does a shape for the object ← subject direction: the description is written about the property, so on the inverse path it would say the opposite of what the shape holds.

Every property shape carries shui:showAlways, and a shape for the object ← subject direction carries shui:inversePath as well. A node shape carries the foaf:depiction of its target class wherever the data graph or the class’s own vocabulary offers one. The catalog itself records the data graph it was generated from, when it was written, and the classes it manages.

Caveats¤

Names, labels and descriptions are requested in English. Where a vocabulary offers no English text, Corporate Memory answers in whatever language it does have, and the literal carries that language rather than claiming to be English. A class or property the deployment knows nothing about falls back to a name built from its IRI, which has no language at all and is written without a tag.

A property used by several classes gets one property shape, shared by every node shape that uses it. Its sh:nodeKind is decided by whichever of those classes sorts first, so a property carrying IRI values under one class and literal values under another is described as only one of the two - the same one on every run over the same data.

A class or property that is a blank node is skipped. Instance data sharing a graph with an ontology types things with anonymous class expressions, and a shape has nothing to target in one.

sh:datatype rdf:langString is added as soon as any value of a property carries a language tag, however few of them do. A property whose values mix tagged and untagged literals therefore gets a shape that its own source data does not satisfy.

Adding to an existing catalog replaces the name, label and description of every shape the run writes, and the classes the catalog declares, rather than leaving an earlier run’s beside them. Everything else in the catalog is left alone, including shapes this run did not produce - that is what adding is for.

Example¤

Given a data graph typing instances with a vocabulary the deployment knows:

ex:Person123 a ex:Person ;
    ex:name "John" ;
    ex:knows ex:Person456 .

the task generates a node shape for ex:Person

graph:90ee6e27-59b1-5ac8-9d7a-116c60c6791a a sh:NodeShape ;
  rdfs:label "Person (ex:)"@en ;
  sh:name "Person (ex:)"@en ;
  sh:description "A person."@en ;
  sh:property
    graph:0fcf371d-f99a-5eeb-ab50-6e6b5fbb0e06 ,
    graph:dd5c6728-75a2-5215-8a5d-f9cd4077aaea ;
  sh:targetClass ex:Person .

together with a property shape for each of ex:knows and ex:name, of which the first reads

graph:0fcf371d-f99a-5eeb-ab50-6e6b5fbb0e06 a sh:PropertyShape ;
  rdfs:label "knows (ex:)"@en ;
  sh:name "knows (ex:)"@en ;
  sh:nodeKind sh:IRI ;
  sh:path ex:knows ;
  shui:showAlways true .

Parameter¤

Input data graph¤

The knowledge graph holding the instance data to analyze.

• ID: data_graph_iri
• Datatype: scheme:string
• Default Value: None

Output shape catalog¤

The graph the generated shapes are written to.

• ID: shapes_graph_iri
• Datatype: scheme:string
• Default Value: None

Handle an existing shape catalog¤

What to do when the output shape catalog already exists.

• ID: existing_graph
• Datatype: string
• Default Value: stop

Import into the central shape catalog¤

If enabled, the generated catalog is imported into the central shape catalog by adding an owl:imports statement to it. Shapes in a catalog the central one does not import are not picked up.

• ID: import_shapes
• Datatype: boolean
• Default Value: false

Advanced Parameter¤

Output shape catalog label¤

The label of the shape catalog. Left empty, a new catalog gets a generated label, and a catalog being added to keeps the label it has - but a catalog being replaced is rewritten whole, generated label included. Only a label tagged “en” or carrying no language tag counts as an existing label, and only such a label is replaced.

• ID: label
• Datatype: string
• Default Value: None

Fetch namespace prefixes from prefix.cc¤

If enabled, the namespace prefix list is fetched from https://prefix.cc instead of the local prefix database, falling back to the local database when the service cannot be reached. Prefixes defined in the project win over both. Enabling this exposes your IP address to prefix.cc; no other data is shared. If unsure, leave it disabled. See https://prefix.cc/about.

• ID: prefix_cc
• Datatype: boolean
• Default Value: false

Properties to ignore¤

The properties to leave out of the generated shapes, as IRIs, one per line.

• ID: ignore_properties
• Datatype: multiline string
• Default Value: http://www.w3.org/1999/02/22-rdf-syntax-ns#type

Classes to ignore¤

The classes to leave out of the generated shapes, as IRIs, one per line.

• ID: ignore_types
• Datatype: multiline string
• Default Value: None

Include task provenance¤

If enabled, the shape catalog also records the task that generated it and the parameter values it ran with. Where the type of the task cannot be determined, nothing is recorded and the run continues with a warning in the log.

• ID: plugin_provenance
• Datatype: boolean
• Default Value: false

Omit the namespace prefix from property names¤

If enabled, property shape names and labels leave off the trailing namespace prefix, reading “label” rather than “label (rdfs:)”. Node shape names always keep it.

• ID: omit_namespace_addon
• Datatype: boolean
• Default Value: false

Declare the classes the catalog manages¤

If enabled, the shape catalog states with shui:managedClasses which classes it manages: node shapes, property shapes, property groups, SPARQL constraints, prefix declarations, chart visualizations, table reports, widget integrations and workflow triggers.

• ID: managed_classes
• Datatype: boolean
• Default Value: true

Add depictions to node shapes¤

If enabled, a node shape is given the foaf:depiction of its target class, where one can be found. The class is looked up in the data graph and in the graph named by its own namespace, tried both with and without the trailing separator, because the graph a vocabulary was loaded into is not recorded anywhere. A class depicted more than once contributes one of its depictions, the same one on every run.

• ID: depictions
• Datatype: boolean
• Default Value: true

Comments