Translating the ArchiMate Model Exchange File Format to RDF
The Open Group has published the ArchiMate Model Exchange File format - a standard file format for the exchange of ArchiMate models between different tools. Resources can be found at: https://www.opengroup.org/open-group-archimate-model-exchange-file-format
The aim for this project, archimate2rdf, is to create a translator that fully translates a ArchiMate exchange file to a corresponding OWL ontology.
Current status:
- It works :-)
- All standard and custom properties and relationships are translated, even relationships from or to relationships (Archimate 3.0)
- Only model elements and relationships are translated, diagrams are currently not supported;
- Target ontology is work-in-progress, some design decisions are not final, comments are welcomed!
We expect the ontology to change. Our current focus is the creation of a complete translator. This means that the current ontology is as close to the XSD as possible, focussing on simplicity. The ontology doesn't contain any "knowledge" about the ArchiMate language that is not available in the XSD (the ontology is actually generated from the XSD at this moment!). We expect that the final ontology will have a more semantic orientation.
Build, or download archimate2rdf.jar
java -jar archimate2rdf.jar <input.xml> <output.xml>
The ontology we use can be found here: https://bp4mc2.org/def/archimate
Unfortunately, no official archimate ontology exists. Different initiatives have been undertaken (listed below - as far as we know), but all these initiatives have in fact only a transformation goal. And for that, you need a target ontology! Our own target ontology is nothing better: something we need to translate a particular archimate exchange file to. Interchange of RDF archimate files will only be possible if we all agree upon one, canonical, ArchiMate ontology.
We urge the Open Group to publish the archimate ontology at their own namespace! We welcome them to use ours, or one of the other ontologies mentioned below. Please :-)
Some other people have tried to create an archimate ontology, or mention the need for one:
- https://www.linkedin.com/pulse/from-archimate-language-web-ontology-dr-nicolas-figay
- https://github.com/nfigay/archimate.owl
- https://github.com/archimatetool/OWLExchange
- https://forum.archimatetool.com/index.php?topic=309.0
- https://groups.google.com/forum/#!topic/open-archimate-forum/poUR2BRqve0
- https://github.com/ikm-group/ArchiMEO/blob/master/ARCHIMEO/ARCHIMATE/ArchiMate.ttl
- https://link.springer.com/chapter/10.1007/978-3-642-54734-8_7
- https://www.sba-research.org/wp-content/uploads/publications/ant_csimq14.pdf
We used the ArchiMate Model Exchange File format to create the transformation. The reason we used this file format is:
- It's a standard - a clear meaning is understood;
- It's tool independent - anybody can use our transformation software, regardless of the tool used;
- It's an XML format, so transformation using XSL is quite straightforward and easily understood.
The actual transformation XSL can be found here: src/main/java/resources/xsl/archimate2rdf.xsl
The following rules were followed for the transformation:
An ArchiMate Element is mapped to an OWL Class.
The xsi:type of the Element is used as the fragment part for the URI of the OWL Class, UpperCamelCase.
<element xsi:type="BusinessRole">is mapped toarchimate:BusinessRole
An ArchiMate Relationship is mapped to an OWL ObjectProperty.
The xsi:type of the Relationship is used as the fragment part for the URI of the OWL ObjectProperty, lowerCamelCase.
<relationship xsi:type="Assignment">is mapped toarchimate:assignment
An alternative to the stricted naming might be to use the verbalisation, for example: "isAssignedTo". Because this would introduce a more complex mapping, we have not done that (yet).
The Access relationship is a relationship that is further subtyped using an accessType property. The most natural way of mapping this to an ontology, is to use subproperties.
<relationship xsi:type="Access" accessType="Read">is mapped toarchimate:readAccess
The name of an element is mapped to rdfs:label. You could argue that a archimate-specific property would be more appropriate. From a interoperability perspective, we prefer all resources to have a rdfs:label.
The documentation of an element is mapped to rdfs:comment. You could argue that a archimate-specific property would be more appropriate. From a interoperability perspective, we prefer rdfs:comment.
In some cases, an Archimate model will have relationships with their own properties:
- A relationship can have a name;
- A relationship can have documentation;
- It is possible to relate an element to a relationship (as of ArchiMate 3.0);
- It is possible to relate a relationship to an element (as of ArchiMate 3.0).
As is described above, relationships are modelled as OWL ObjectProperties. As such, it is not possible to add extra data to these relationships, because the triple itself doesn't have a URI. For example, the example below depicts how a ArchiMate relationship is mapped to RDF:
ex:AC1 a archimate:ApplicationComponent;
rdfs:label "Application Component #1";
archimate:flow ex:AC2;
.
ex:AC2 a archimate:ApplicationComponent;
rdfs:label "Application Component #2"@en;
.
It is clearly visible, that even in the pre-3.0 versions of ArchiMate, this mapping is problematic with regard to the name or documentation of the relationship. An how would we map the following model:
The solution is the use of rdf:Statement. From this statement, the original triple could even be inferred. We prefer, however, to explicitly include the original triple for easy of use, for example in SPARQL queries:
ex:AC1 a archimate:ApplicationComponent;
rdfs:label "Application Component #1"@en;
archimate:flow ex:AC2;
.
ex:AC2 a archimate:ApplicationComponent;
rdfs:label "Application Component #2"@en;
.
ex:DO1 a archimate:DataObject;
rdfs:label "Message"@en;
.
ex:rel1 a rdf:Statement;
rdf:subject ex:AC1;
rdf:object ex:AC2;
rdf:predicate archimate:flow;
archimate:association ex:DO1;
.
The example above used human-readable URI's for the example resources (ex:AC1, ex:AC2, ex:DO1 and ex:rel1). The actual mapping uses the original identifier from the XML file. As these identifiers are UUID's, a standard mapping is available to create URI's (more specific: URN's) from these UUID's.
<element identifier="93bf197b-5a00-4b44-856c-2638ae4e30fd">is mapped to<urn:uuid:93bf197b-5a00-4b44-856c-2638ae4e30fd>
It could be argued that we should use URL's instead of URN's. The converter might have an option to mint such URI's. At this moment, we've sticked with the more straightforward mapping from UUID to URN.
Custom properties are mapped to new properties. This means that adding custom properties to the model will extend the ArchiMate ontology. The URI's for these properties are created from the identifier of the propertydefinition and the name of the model:
<property propertyDefinitionRef="propid-1">is mapped to<http://bp4mc2.org/archimate/my-model/property/propid-1> rdfs:subClassOf archimate:property.(With "My model" being the name of this specific model)
From this, the actual use of such a property can be mapped to this newly created property:
<property propertyDefinitionRef="propid-1"><value xml:lang="en">Active Structure</value></property>is mapped to@prefix property: <http://bp4mc2.org/archimate/my-model/property/>.[] property:propid-1 "Active Structure."(where[]should be replaced with the URI of the specific model element)
It could be argued that we shouldn't use the identifier of the property, but the more readable label of the property. Because this would add more complexity to the translator, we have not done that (yet).