This manuscript introduces J2RM : a tool to process mappings from JSON data to RDF triples guided by an OWL2 ontology structure. The mappings are de ned as annotation properties associated with each ontology entity of interest. They are embedded in an ontology le so that they can be readily deployed and shared to automate RDF-graph creation. In this paper, we motivate the need for such mappings, describe some of their de nitions on a use case example, present the formal grammar of the mapping language, and explain how these mappings work. We conclude with a discussion of the key aspects, main contributions, and future improvements.
Quite often data transformation tasks consume a lot of engineering e ort when
dealing with heterogeneous data models and formats. Speci cally, creating an
RDF-graph based on data extracted from a closed and proprietary information
system can be a daunting task. A simple approach to extract the required and
curated data from these systems is to expose the data in an \easy-to-process"
format, usually, JSON, as an intermediary representation. JSON has been used
extensively in a variety of processing tasks as a serialization format becoming the
universal format for data interchange on the Web [
RDF-graph creations from JSON data following an OWL2 ontology structure.
The mappings are declared as annotation properties associated with each
ontology entity of interest6. The mappings are embedded in an ontology le so they
can be readily deployed to automate the graph creation from a \standardized"
JSON structure, tailored from any information systems' data (see Figure 1).
With J2RM, one could work with di erent JSON structures where all mappings
are embedded in a speci c ontology le. Some transformation and mapping
languages have been proposed to generate RDF from non-RDF data, including
SPARQL-Generate [
Information System (closed and proprietary)
JSON (as an intermediary format)
J2RM Processor
J2RM Mappings <_ _ _> <_ _ _> 6 The IRI used to identify the mappings is de ned in the J2RM con guration le. 7 Although the RML mappings may be connected to an ontology, these are de ned in a separate de nition le. } ], "team": [ { "ind_id": "401636", "role": "CIA",
"name": "Dr Susan Storm", erated8. In this case, the meta-character \#" indicates that the mapped value is used \as is" in the IRI9. The meta-character \!" in #3 indicates that the string value (with blank spaces) is used to generate an IRI (with replacements): d:Area#Clinical-Medicine a m:Area. #4 maps to an array formed of composite values based on the tree structure: ["A19453-401636", "A19453-401443"], which are used to generate two instances of m:ChiefAnalyst.
Datatype (dp) and annotation (ap) prop. mappings: create a triple for each mapped value with the structure <data#ID> <dp|ap> "value"^^<xsd:type>. J2RM analyzes the ontology; for each class (and sub-classes) that has <dp|ap> as a class restriction, it will create a triple for each mapped instance. One example of #5 is d:Analyst#401636 m:fullName "Dr Susan Storm"^^xsd:string considering that m:Analyst has m:fullName in its class restrictions. In this case, the meta-character \~" indicates that the mapped value is used to automatically 8 Empty mapped values (\", null, {or not existent{) are not processed. 9 The created triple is <https://orcid.org/0X30-01X1-68X0-083X> a m:ORCID create an rdfs:label triple as well (#13 presents similar examples)10. #6 creates a triple for each value found when splitting the mapped values using the delimiter \ | " and, thus, it will generate three keywords. #7 de nes a \conditional path": in this case, it will create a triple with the mapped value of \4.91" because the restriction (expression after meta-character \%") evaluates to true: the scheme and year values are mapped and evaluated correctly. In #8, the meta-character \<" de nes a mapping to a common JSONObject ancestor: for the m:Grant class with instances mapped as doc/publicData/id, the ancestor is publicData11.
Object prop. mappings: create triples between sets of mapped values for each identi ed class that is applicable in the analyzed context (class hierarchies, sub-properties, etc). The structure generated is <domainData#ID> <op> <rangeData#ID>, where <domainData#ID> correspond to the mapped instances of each <op> domain class, and <rangeData#ID> correspond to the mapped instances of each <op> range class. The mappings are paths that de ne the connection between <domainData#ID> and <rangeData#ID>. Simple cases, such as #912 and #10, nd the connection between the instances in a single path: in #9, /doc connects the domain instances /doc/id="A19453" with the range instances /doc/FoR code="12908", creating the triple d:GrantApp#A19453 m:hasFoR d:FoR#12908. The meta-character \@" is used (#9, #12, #13) to indicate the entity (domain class) attached to the path (useful for entity disambiguation). In #10, when applying to the domain class m:Analyst, the mapping results in an array of values for both, the domain (["401636", "401443"]) and the range (same as #2). Internally, the tool keeps track of the context for each mapped JSONObject that could result in a valid connection. #11 illustrates a mapping based on two di erent paths: for domain (D=, indicates the usage of the already known instances from the domain classes) and range (R=..., indicates the mapping to the values that are equal to "CIA"). #12 illustrates two mappings: one where explicitly disambiguate the domain and range classes to use (CV->FoR), and other, 10 rdfs:label creation might be useful in some graph search and visualization tools. 11 The created triple is d:Grant#19453 m:link "http://test.com/key/19453"^^xsd:anyURI 12 #9 de nes two mappings that apply to distinct domain classes. <path 1>|=|<path 2>, where it will map to values of <path 1> only if those values are equal to values of <path 2>.
Along with each mapping, one can specify the target endpoint and graph. Target endpoint is a label that identi es a SPARQL endpoint access13 where the triples will be created. Examples: test, prod. Target graph is the named graph where the triples will be created. It is de ned as a namespace pre x in the ontology le. Examples: g0-testing, g0-prod. The namespace pre x IRI will be used as the named graph for the triple creation for that speci c mapping. 3
J2RM gives information architects a simple mechanism to de ne the necessary mapping rules for an automated RDF-graph creation task guided by an OWL2 ontology structure from any JSON data. The key aspect is that the mappings are embedded in an ontology le: this does not imply that the JSON structure is intrinsically tied to the OWL2 model. For di erent JSON structures, one could de ne each type of mappings in di erent ontology les. J2RM is in its early development stages. It has been tested on three di erent domain ontologies. We will increase the support for more complex JSON mappings and more OWL2 axioms. The major contributions of this tool are: the ability to selectively extract data and perform basic operations on the source JSON structure, the \portability" of the mappings embedded in the OWL2 ontology le as annotation properties attached to classes and properties, and its ease of use while hiding the complexity of creating RDF triples following OWL2 axioms.