JOPA is a Java OWL persistence framework aimed at efficient programmatic access to OWL2 ontologies and RDF graphs in Java. The system is based on integrity constraints [1] in OWL that JOPA uses to establish a contract between a JOPA-enabled Java application and an OWL ontology. Note, however, that for practical purposes of working with triple stores, this OWL integrity constraints-based contract is not required.
The library architecture and API is similar to JPA (see [2]) so that Java developers find it familiar.
Similarly to object-relational mapping (ORM), OOM enables to map ontological constructs to constructs of an object-oriented programming language and vice versa.
More specifically, OOM in JOPA maps (using the JLS [3] terminology):
Ontology | OO Language |
---|---|
OWL Class | Reference type |
Object property | Reference type member |
Data property | Primitive type member (+ String, Date, MultilingualString) |
Annotation property | Reference or primitive type member |
Class assertions | Reference type instance or @Types record |
All this means that individuals belonging to an OWL class can be retrieved as instances of a (Java) class.
Note: OOM works also for RDFS ontologies. See the wiki for details.
Here is a simple example of a JOPA entity:
@Namespace(prefix = "skos", namespace="http://www.w3.org/2004/02/skos/core#")
@OWLClass(iri = "skos:Concept")
public class Term {
@Id
private URI id;
@OWLAnnotationProperty(iri = "skos:prefLabel")
private MultilingualString label;
@OWLAnnotationProperty(iri = "skos:definition")
private MultilingualString definition;
@OWLObjectProperty(iri = "skos:narrower")
private Set<Term> children;
@Types
private Set<String> types;
@Properties
private Map<String, Set<String>> properties;
}
A member annotated with the @Inferred
annotation represents a field whose values are retrieved using a reasoner. As such,
they can, for example, contain values of an inverse object property (like in the Jedi example).
There are limitations to this: JOPA requires explicit class assertion to be able to load individual as instance of a class. And, inferred values are read-only. These restrictions have pragmatic reasons - if the knowledge is inferred, it cannot be directly modified/removed, so attempting to remove an inferred value does not have any direct effects.
OOM is not meant to completely capture the ontological model. It would not even make much sense. One of the main features
of JOPA is its ability to work with knowledge which is not part of the object model. This is done using members annotated
with @Types
and @Properties
. @Types
field contains all OWL classes whose instance the particular individual represented by
an object is (except the one mapped by the object's Java class).
@Properties
field contains values of properties not mapped by object model. This way, the application gets (although limited)
access to unmapped property values (e.g. values of newly added properties), without the need to adjust the object model and recompile.
JOPA supports object-level transactions. In addition, it makes transactional change visible to the transaction that made them. This means that when you add an instance of some class during a transaction and then list all instances of that class (during the same transaction), you'll see the newly added instance as well.
There are limitations to this approach. Currently, pending changes are not taken into account when doing inference. Also, the current version of RDF4J OntoDriver is not able to include pending changes into results of SPARQL queries.
Similarly to JPA and JDBC driver, JOPA sits on top of an OntoDriver instance, which provides access to the underlying storage. There are two main reasons for such a split - first, it decouples storage-specific API usage from the more generic OOM core. Second, it enables the application to switch the underlying storage with as little as 2-3 lines of configuration code. Nothing else needs to be modified.
Supported storages:
The whole framework consists of several modules:
Other modules represent integration tests and various utilities.
Check out the Wiki for general information about JOPA, explanation of its features and their usage. The content is being gradually created and updated.
Javadoc of the latest published version is available at https://kbss.felk.cvut.cz/jenkins/job/jopa-stable/javadoc/index.html?overview-summary.html.
For practical examples of JOPA features, see the JOPA examples repository.
JOPA examples can be found in a separate repository at https://github.com/kbss-cvut/jopa-examples.
A real-world, up-to-date project using JOPA is TermIt - a SKOS-compatible vocabulary manager.
Note that JOPA requires Java 17 or later.
There are two ways of getting JOPA for a project:
Basically, the jopa-impl module and one of the OntoDriver implementations is all that is needed:
<dependencies>
<dependency>
<groupId>cz.cvut.kbss.jopa</groupId>
<artifactId>jopa-impl</artifactId>
</dependency>
<dependency>
<groupId>cz.cvut.kbss.jopa</groupId>
<artifactId>ontodriver-rdf4j</artifactId>
<!-- OR <artifactId>ontodriver-jena</artifactId> -->
<!-- OR <artifactId>ontodriver-owlapi</artifactId> -->
</dependency>
</dependencies>
More information about JOPA can be found, for example, in articles [4], [5], [6] and on the GitHub Wiki.
JOPA build status and code metrics can be found at:
A performance comparison of JOPA and other object-triple mapping libraries can be found at https://kbss.felk.cvut.cz/web/otm-benchmark.
A comprehensive comparison - feature and performance - of object-triple mapping libraries is presented in [7].
Some related libraries:
@Transactional
annotation) with JOPA.Notable changes:
jopa-api
module etc.IN
, NOT LIKE
, <>
operators in SOQL.See CHANGELOG.md for detailed change history.
LGPLv3