Jena Ontology API

In this section I will give a general introduction to the the Jena2 ontology API. I'll also describe a range of common user tasks. I won't go into all of the many details of the API here: you should expect to refer to the Jena2 Javadoc to get full details of the capabilities of the API. Feedback, (via the jena-dev support list) is welcomed.

Prerequisites

I will assume that you have a basic familiarity with RDF and with Jena. There are many other Jena help documents you can read for background on these topics. There is also a Jena programming tutorial you can follow through.

Jena is a programming toolkit, using the Java programming language. While there are a few command-line tools to help you perform some key tasks using Jena, mostly you use Jena by writing Java programs. The examples in this document will be primarily code samples.

I also won't be explaining the OWL or RDFS ontology languages in much detail in this document. You should expect to refer to supporting documentation for details on those languages, for example the W3C OWL document index.

Note: At the time of writing, work on OWL version 1.1 is still underway. No decision has yet been made about when Jena will support the new OWL 1.1 features. I will only use OWL 1.0 features in this document.

Overview

The section of the manual is broken into a number of sections. You do not need to read them in sequence, though later sections may refer to concepts and techniques introduced in earlier sections. The sections are:

1. Overview
2. General concepts
3. Running example: the ESWC ontology
4. Creating ontology models
5. Compound ontology documents and imports processing
6. The generic ontology type: OntResource
7. Ontology classes and basic class expressions
8. Ontology properties
9. More complex class expressions
10. Instances or individuals
11. Ontology meta-data
12. Ontology inference: overview
13. Working with DAML+OIL ontologies
14. Working with persistent ontologies
15. Experimental ontology tools
16. Common ontology application problems and sample programs

Further assistance

I hope that this document will be sufficient to help most readers to get started using the Jena ontology API. For further support, please post questions to the Jena support list: jena-dev@yahoogroups. You can also report bugs directly into the Jena bug tracker on SourceForge.

Please note that we ask that you use the support list or the bug-tracker to communicate with the Jena team, rather than send email to the team members directly. This helps us manage Jena support more effectively, and facilitates contributions from other Jena community members.

General concepts

In a widely-quoted definition, an ontology is

"… a specification of a conceptualization" [Gruber, T. 1993]

Let's unpack that brief characterisation a bit. It means that an ontology allows a programmer to specify, in an open, meaningful, way the concepts and relationships that collectively characterise some domain. Examples might be the concepts of red and white wine, grape varieties, vintage years, wineries and so forth that characterise the domain of 'wine', and relationships such as 'wineries produce wines', 'wines have a year of production'. This wine ontology might be developed initially for a particular application, such as a stock-control system at a wine warehouse. As such, it may be considered similar to a well-defined database schema. The advantage to an ontology is that it is an explicit, first-class description. So having been developed for one purpose, it can be published and reused for other purposes. For example, a given winery may use the wine ontology to link its production schedule to the stock system at the wine warehouse. Alternatively, a wine recommendation program may use the wine ontology, and a description (ontology) of different dishes to recommend wines for a given menu.

There are many ways of writing down an ontology, and a variety of opinions as to what kinds of definition should go in one. In practice, the contents of an ontology are largely driven by the kinds of application it will be used to support. In Jena, we do not take a particular view on the minimal or necessary components of an ontology. Rather, we try to support a variety of common techniques. In this manual I try to explain what is – and to some extent what isn't – possible using Jena's ontology support.

Since Jena is fundamentally an RDF platform, Jena's ontology support is limited to ontology formalisms built on top of RDF. Specifically this means RDFS, the varieties of OWL, and the now-obsolete DAML+OIL. I will provide a very brief introduction to these languages here. Please refer to the extensive on-line documentation for these formalisms for complete and authoritative detail.

RDFS

RDFS is the weakest ontology language supported by Jena. RDFS allows the ontologist to build a simple hierarchy of concepts, and a hierarchy of properties. Consider the following trivial characterisation (with apologies to biology-trained readers!):

Table 1: A simple concept hierarchy

Using RDFS, I can say that my ontology has five classes, and that Plant is a sub-class of Organism and so on. So every animal is also an organism. A good way to think of these classes is as describing sets of individuals: organism is intended to describe a set of living things, some of which are animals (i.e. a sub-set of the set of organisms is the set of animals), and some animals are fish (a subset of the set of all animals is the set of all fish).

To describe the attributes of these classes, I can associate properties with the classes. For example, animals have sensory organs (noses, eyes, etc.). A general property of an animal might be senseOrgan, to denote any given sensory organs a particular animal has. In general, fish have eyes, so a fish might have a eyes property to refer to a description of the particular eye structure of some species. Since eyes are a type of sensory organ, we can capture this relationship between these properties by saying that eye is a sub-property-of senseOrgan. Thus if a given fish has two eyes, it also has two sense organs. (It may have more, but we know that it must have two).

I can describe this simple hierarchy with RDFS. In general, the class hierarchy is a graph rather than a tree (i.e. not like Java class inheritance). The slime mold is popularly, though perhaps not accurately, thought of as an organism that has characteristics of both plants and animals. I might model a slime mold in my ontology as a class that has both plant and animal classes among its super-classes. RDFS is too weak a language to express that a thing cannot be both a plant and an animal (which is perhaps lucky for the slime molds). In RDFS, I can only name the classes, I cannot construct expressions to describe interesting classes. However, for many applications it is sufficient to state the basic vocabulary, and RDFS is perfectly well suited to this.

Note also that I can both describe classes, in general terms, and I can describe particular instances of those classes. So there may be a particular individual Fred who is a Fish (i.e. has rdf:type Fish), and who has two eyes. His companion Freda, a Mexican Tetra, or blind cave fish, has no eyes. One use of an ontology is to allow us to fill-in missing information about individuals. Thus, though it is not stated directly, we can deduce that Fred is also an Animal and an Organism. Assume that there was no rdf:type asserting that Freda is a Fish. We may still infer Freda's rdf:type since Freda has lateral lines as sense organs, and these only occur in fish. In RDFS, we state that the domain of the lateralLines property is the Fish class, so an RDFS reasoner can infer that Freda must be a fish.

OWL

In general, OWL allows me to say everything that RDFS allows, and much more besides. A key part of OWL is the ability to describe classes in more interesting and complex ways. For example, in OWL we can say that Plant and Animal are disjoint classes: no individual can be both a plant and an animal (which would have the unfortunate consequence of making SlimeMold an empty class). SaltwaterFish might be the intersection of Fish and the class SeaDwellers (which also includes, for example, cetaceans and sea plants).

Suppose I have a property covering, intended to represent the scales of a fish or the fur of a mammal. I can now refine the mammal class to be 'animals that have a covering that is hair', using a property restriction to express the condition that property covering has a value from the class Hair. Similarly TropicalFish might be the intersection of the class of Fish and the class of things that have TropicalOcean as their habitat.

Finally (for this brief overview), I can say more about properties in OWL. In RDFS, properties can be related via a property hierarchy. OWL extends this by allowing properties to be denoted as transitive, symmetric or functional, and allow one property to be declared to be the inverse of another. OWL also makes a distinction between properties that have data-values (known as literals in RDF terminology) as their range, or other individuals. Respectively these are datatype properties and object properties. A consequence of the RDF lineage of OWL is that OWL ontologies cannot make statements about literal values. I cannot say in RDF that 7 has the property of being a prime number. I can, of course, say that the class of primes includes 7, but that doesn't require a number to be the subject of an RDF statement. In OWL, this distinction is important since only object properties can be transitive or symmetric.

The OWL language is sub-divided into three syntax classes: OWL Lite, OWL DL and OWL Full. OWL DL does not permit some constructions allowed in OWL Full, and OWL Lite has all the constraints of OWL DL plus some more. The intent for OWL Lite and OWL DL is to make the task of reasoning with expressions in that subset more tractable. Specifically, OWL DL is intended to be able to be processed efficiently by a description logic reasoner. OWL Lite is intended to be amenable to processing by a variety of reasonably simple inference algorithms, though experts in the field have challenged how successfully this has been achieved.

While the OWL standards documents note that OWL builds on top of the (revised) RDF specifications, it is possible to treat OWL as a separate language in its own right, and not something that is built on an RDF foundation. This view uses RDF as a serialisation syntax; the RDF-centric view treats RDF triples as the core of the OWL formalism. While both views are valid, in Jena we take the RDF-centric view.

DAML+OIL

DAML+OIL is very similar to OWL Full. This is not surprising, since the W3C's Web Ontology Working Group, who designed OWL, took DAML+OIL as their starting point. Some constructions in OWL have been added to the capabilities of DAML+OIL, and one or two have been removed. For the purposes of this brief overview, however, the expressiveness of DAML+OIL is comparable to that of OWL.

Note: DAML+OIL is effectively now obsolete. While you can still find some DAML ontologies on the web, you should work with OWL or RDFS to maximise the longevity of your models.

Ontology languages and the Jena Ontology API

As I outlined above, there are various different ontology languages available for representing ontology information on the semantic web. They range from the most expressive, OWL Full, through to the weakest, RDFS. Through the Ontology API, Jena aims to provide a consistent programming interface for ontology application development, independent of which ontology language you are using in your programs.

The Jena Ontology API is language-neutral: the Java class names do not mention the underlying language. For example, the OntClass Java class can represent an OWL class, RDFS class, or DAML class. To represent the differences between the various representations, each of the ontology languages has a profile, which lists the permitted constructs and the names of the classes and properties. Thus in the DAML profile, the URI for object property is daml:ObjectProperty (short for http://www.daml.org/2001/03/daml+oil#ObjectProperty), in the OWL profile is it owl:ObjectProperty (short for http://www.w3.org/2002/07/owl#ObjectProperty) and in the RDFS profile it is null since RDFS does not define object properties.

The profile is bound to an ontology model, which is an extended version of Jena's Model class. The base Model allows access to the statements in a collection of RDF data. OntModel extends this by adding support for the kinds of objects expected to be in an ontology: classes (in a class hierarchy), properties (in a property hierarchy) and individuals. When you're working with an ontology in Jena, all of the state information remains encoded as RDF triples (which Jena calls Statements) stored in the RDF model. The ontology API doesn't change the RDF representation of ontologies. What it does do is add a set of convenience classes and methods that make it easier for you to write programs that manipulate the RDF statements.

The predicate names defined in the ontology language correspond to the accessor methods on the Java classes in the API. For example, an OntClass has a method to list its super-classes, which corresponds to the values of the subClassOf property in the RDF representation. This point is worth re-emphasising: no information is stored in the OntClass object itself. When you call the OntClass listSuperClasses() method, Jena will retrieve the information from the underlying RDF statements. Similarly adding a subclass to an OntClass asserts an additional RDF statement into the model.

Ontologies and reasoning

One of the main reasons for building an ontology-based application is to use a reasoner to derive additional truths about the concepts you are modelling. We saw a simple instance of this above: the assertion "Fred is a Fish" entails the deduction "Fred is an Animal". There are many different styles of automated reasoner, and very many different reasoning algorithms. Jena includes support for a variety of reasoners through the inference API. A common feature of Jena reasoners is that they create a new RDF model which appears to contain the triples that are derived from reasoning, as well as the triples that were asserted in the base model. This extended model is, nevertheless, still conforms to the contract for Jena models. So it can be used wherever a base model can be used. The ontology API exploits this feature: the convenience methods the ontology API provides can query an extended inference model in just the same way as a plain RDF model. In fact, this is such a common pattern that we provide simple recipes for constructing ontology models whose language, storage model and reasoning engine can all be simply specified when an OntModel is created. I'll show examples later.

Figure 2 shows one way of visualising this:

Figure 2: the statements seen by the OntModel

Graph is an internal Jena interface that supports the composition of sets of RDF triples. The asserted statements, which may have been read in from an ontology document, are held in the base graph. The reasoner, or inference engine, can use the contents of the base graph and the semantic rules of the language to show a more complete set of statements. This is also presented via a Graph interface, so the model works only with the outermost interface. This regularity allows us to very easily build ontology models with or without a reasoner. It also means that the base graph can be an in-memory store, a database-backed persistent store, or some other storage structure altogether (e.g. an LDAP directory) again without affecting the operation of the ontology model.

RDF-level polymorphism and Java

Deciding which Java abstract class to use to represent a given RDF resource can be surprisingly subtle. Consider the following RDF sample:

<owl:Class rdf:ID="DigitalCamera">
</owl:Class>

This declares that the resource with the relative URI #DigitalCamera is an OWL ontology class. It suggests that it would be appropriate to model that declaration in Java with an instance of an OntClass. Now suppose I add a triple to the RDF model to augment the class declaration with some more information:

<owl:Class rdf:ID="DigitalCamera">
  <rdf:type owl:Restriction />
</owl:Class>

Now I am saying that #DigitalCamera is an OWL Restriction. Restriction is a subclass of owl:Class, so this is a perfectly consistent operation. The problem I have is that Java does not allow me to dynamically change the Java class of the object representing this resource. The resource has not changed: it still has URI #DigitalCamera. But the appropriate Java class Jena might choose to model it has changed from OntClass to Restriction. Conversely, if I subsequently remove the rdf:type owl:Restriction from the model, using the Restriction Java class is no longer appropriate.

Even worse, OWL Full allows me to state the following (rather counter-intuitive) construction:

<owl:Class rdf:ID="DigitalCamera">
  <rdf:type owl:ObjectProperty />
</owl:Class>

That is, #DigitalCamera is both a class and a property. While this may not be a very useful operation, it illustrates a basic point: we cannot rely on a consistent or unique mapping between an RDF resource and the appropriate Java abstraction.

Jena accepts this basic characteristic of polymorphism at the RDF level by considering that the Java abstraction (OntClass, Restriction, DatatypeProperty, etc.) is just a view or facet of the resource. That is, there is a one-to-many mapping from a resource to the facets that the resource can present. If the resource is typed as an owl:Class, it can present the OntClass facet; given other types, it can present other facets. Jena provides the .as() method to efficiently map from an RDF object to one of its allowable facets. Given a RDF object (i.e. an instance of com.hp.hpl.jena.rdf.model.RDFNode or one of its sub-types), I can get a facet by invoking as() with an argument that denotes the facet required. Specifically, the facet is identified by the Java class object of the desired facet. For example, to get the OntClass facet of a resource, I can do:

Resource r = myModel.getResource( myNS + "DigitalCamera" );
OntClass cls = (OntClass) r.as( OntClass.class );

This pattern allows me to defer decisions about the correct Java abstraction to use until run-time. The choice can depend on the properties of the resource itself. If a given RDFNode will not support the conversion to a given facet, it will raise a ConversionException. I can test if .as() will succeed for a given facet with canAs(). This RDF-level polymorphism is used extensively in the Jena ontology API to allow maximum flexibility in handling ontology data.

Running example: the ESWC ontology

To illustrate the principles of using the ontology API, I will use examples drawn from Tom Heath's ESWC ontology. This ontology presents a simple model for describing the concepts and activities associated with a typical academic conference. A copy of the ontology serialized in RDF/XML is included with the Jena download, see: [eswc-2006-09-21.rdf] (note that you may need to view the page source in some browsers to see the XML code).

A subset of the classes and properties from the ontology are shown in Figure 3:

Figure 3: Classes and properties from ESWC ontology

I will use elements from this ontology to illustrate the ontology API throughout the rest of this document.

Creating ontology models

An ontology model is an extension of the Jena RDF model that provides extra capabilities for handling ontologies. Ontology models are created through the Jena ModelFactory. The simplest way to create an ontology model is as follows:

OntModel m = ModelFactory.createOntologyModel();

This will create an ontology model with the default settings, which are set for maximum compatibility with the previous version of Jena. These defaults are:

• OWL-Full language
• in-memory storage
• RDFS inference, which principally produces entailments from the sub-class and sub-property hierarchies.

In many applications, such as driving a GUI, RDFS inference is too strong (for example, every class is inferred to be a sub-class of owl:Thing. In other applications, stronger reasoning is needed. In general, to create an OntModel with a particular reasoner or language profile, you should pass a model specification to the createOntologyModel call. For example, an OWL model that performs no reasoning at all can be created with:

OntModel m = ModelFactory.createOntologyModel( OntModelSpec.OWL_MEM );

To create an ontology model for a particular language, but leaving all of the other values as defaults, you should pass the URI of the ontology language to the model factory. The URI strings for the various language profiles are:

These URI's are used to look-up the language profile from the ProfileRegistry. Fortunately, the profile registry contains declared constants so that you do not have to remember these URI's. Please note that the URI's denoting OWL Lite and OWL DL are not officially sanctioned by the OWL standard.

To create an ontology model for handling DAML ontologies, use either of:

OntModel m = ModelFactory.createOntologyModel( OntModelSpec.DAML_MEM );
OntModel m = ModelFactory.createOntologyModel( ProfileRegistry.DAML_LANG );

Beyond these basic choices, the complexities of configuring an ontology model are wrapped up in a recipe object called OntModelSpec. This specification allows complete control over the configuration choices for the ontology model, including the language profile in use, the reasoner, and the means of handling compound documents. A number of common recipes are pre-declared as constants in OntModelSpec, and listed below.

For details of reasoner capabilities, please see the inference documentation and the Javadoc for OntModelSpec. See also further discussion below.

Note: it is primarily the choice of reasoner, rather than the choice of language profile, that determines which entailments are seen by the ontology model. However, using an OWL reasoner with DAML source data will result in few additional entailments being seen by the ontology model.

To create a model with a given specification, you should invoke the ModelFactory as follows:

OntModel m = ModelFactory.createOntologyModel( <model spec> );

for example:

OntModel m = ModelFactory.createOntologyModel( OntModelSpec.OWL_MEM_MICRO_RULE_INF );

To create a custom model specification, you can create a new one from its constructor, and call the various setter methods to set the appropriate values. More often, you want a variation on an existing recipe. In this case, you copy an existing specification and then update the copy as necessary:

OntModelSpec s = new OntModelSpec( OntModelSpec.OWL_MEM );
s.setDocumentManager( myDocMgr );
OntModel m = ModelFactory.createOntologyModel( s );

Compound ontology documents and imports processing

The OWL and DAML ontology languages include some facilities for creating modular ontologies that can be re-used in a similar manner to software modules. In particular, one ontology can import another. Jena helps ontology developers to work with modular ontologies by automatically handling the imports statements in ontology models.

The key idea is that the base model of an ontology model is actually a collection of models, one per imported model. That means we have to modify figure 2 a bit. Figure 4 shows how the ontology model builds a collection of import models:

Figure 4: ontology model compound document structure for imports

I will use the term document to describe an ontology serialized in some transport syntax, such as RDF/XML or N3. This terminology isn't used by the OWL or RDFS standards, but it is a convenient way to refer to the written artifacts. However, from a broad view of the interlinked semantic web, a document view imposes artificial boundaries between regions of the global web of data.

We will load an ontology document into an ontology model in the same way as a normal Jena model, using the read method. There are several variants on read, that handle differences in the source of the document (to be read from a resolvable URL or directly from an input stream or reader), the base URI that will resolve any relative URI's in the source document, and the serialisation language. In summary, these variants are:

read( String url )
read( Reader reader, String base )
read( InputStream reader, String base )
read( String url, String lang )
read( Reader reader, String base, String Lang )
read( InputStream reader, String base, String Lang )

You can use any of these methods to load an ontology document. Note that we advise that you avoid the the read() variants that accept a java.io.Reader argument when loading XML documents containing internationalised character sets, since the handling of character encoding by the Reader and by XML parsers is not compatible.

By default, when an ontology model reads an ontology document, it will also locate and load the document's imports. An OWL or DAML document may contain an individual of class Ontology, which contains meta-data about that document itself. For example:

<owl:Ontology rdf:about="">
  <dc:creator rdf:value="Ian Dickinson" />
  <owl:imports rdf:resource="http://jena.hpl.hp.com/example-ont" />
</owl:Ontology>

The construct rdf:about="" is a relative URI that will resolve to the document's base URI: in other words it's a shorthand way of referring to the document itself. The owl:imports line states that this ontology is constructed using classes, properties and individuals from the referenced ontology. When OntModel reads this document, it will notice the owl:imports line and attempt to load the imported ontology into a sub-model of the ontology model being built. The definitions from both the base ontology and all of the imports will be visible to the reasoner.

Each imported ontology document is held in a separate graph structure. This is important: I want to keep the original source ontology separate from the imports. When I write the model out again, normally only the base model is written (the alternative is that all you see is a confusing union of everything). And when I update the model, only the base model changes.

Imports are processed recursively, so if my base document imports ontology A, and A imports B, I will end up with the structure shown in Figure 4. Note that the imports have been flattened out. A cycle check is used to prevent the document handler getting stuck if, for example, A imports B which imports A!

The ontology document manager

Each ontology model has an associated document manager that assists with the processing and handling of documents-related concerns. For convenience, there is one global document manager that is used by default by ontology models. You can get a reference to this shared instance through OntDocumentManager.getInstance(). In many cases, it will be sufficient to simply change the settings on the global document manager to suit your application's needs. However, for more fine-grain control, you can create separate document managers, and pass them to the ontology model when it is created through the model factory. To do this, create an ontology specification object (see above), and set the document manager. For example:

OntDocumentManager mgr = new OntDocumentManager();
// set the mgr's properties now
 … some code …
// now use it
OntModelSpec s = new OntModelSpec( OntModelSpec.RDFS_MEM );
s.setDocumentManager( mgr );
OntModel m = ModelFactory.createOntologyModel( s );

Note that the model retains a reference to the document manager it was created with. Thus if you change a document manager's properties, it will affect models that have previously been constructed with that document manager.

Document manager policy

Since the document manager has a large number of configurable options, there are two ways in which you can customise it to your application requirements. Firstly, you can set the individual parameters of the document manager by Java code. Alternatively, when a given document manager is created it can load values for the various parameters from a policy file, expressed in RDF. The document manager has a list of URL's which it will search for a policy document. It will stop at the first entry on the list that resolves to a retrievable document. The default search path for the policy is: file:./etc/ont-policy.rdf;file:ont-policy.rdf. You can find the default policy, which can serve as a template for defining your own policies, in the etc/ directory under the Jena download directory.

You can set the general properties of the document manager in the policy as follows:

<DocumentManagerPolicy>
  <!-- policy for controlling the document manager's behaviour -->
  <processImports rdf:datatype="&xsd;boolean">true</processImports>
  <cacheModels rdf:datatype="&xsd;boolean">true</cacheModels>
</DocumentManagerPolicy>

You can find the simple schema that declares the various properties that you can use in such an ontology document policy in the vocabularies directory of the Jena download. It's called ont-manager.rdf. To change the search path that the document manager will use to initialise itself, you can either pass the new search path as a string when creating a new document manager object, or call the method setMetadataSearchPath().

The ModelMaker: creating storage on demand

In order for the document manager to build the union of the imported documents (which we may refer to as the imports closure), there must be some means of creating new graphs to store the imported ontologies. Loading a new import means that there needs to be a new graph added Jena defines a model maker as a simple interface that allows different kinds of model storage (in memory, file-backed, in a persistent database, etc.) to be created on demand. For the database case, this may include passing the database user-name and password and other connection parameters. New model makers can be created with the ModelFactory.

There are two cases in which we may want to create storage for models on-demand. The first is when creating the OntModel for the first time, some variants of createOntologyModel will allocate space for the base model (instead of, for example, being handed a base model to use as one of the method arguments). The second case when storage must be allocated is when adding an imported document to the union of imports. These cases often require different policies, so the OntModelSpec contains two model maker parameters: the base model maker and imports model maker, available via getBaseModelMaker() and getImportsModelMaker() methods respectively.

The default specifications in OntModelSpec that begin MEM_ use an in-memory model maker for the both the base model and the imported documents.

Implementation note: internally to Jena, we use Graph as a primary data structure. However, your code will almost always refer to models, not graphs. What's happening is that a Model is a wrapper around the Graph, which balances a rich, convenient programming interface (Model) with a simple, manageable internal data structure (Graph). Hence some potential confusion in that Figure 4, above, refers to a structure containing graphs, but we use a ModelMaker to generate new stores. The document manager extracts the appropriate graph from the containing model. Except in cases where you are extending Jena's internal structures, you should think of Model as the container of RDF and ontology data.

Controlling imports processing

Normally, loading imports during the read() call automatic. To read() an ontology without building the imports closure, call the method setProcessImports( false ) on the document manager object before calling read(). Alternatively, you can set the processImports property in the policy file. You can also be more selective, and ignore only certain URI's when loading the imported documents. To selectively skip certain named imports, call the method addIgnoreImport( String uri ) on the document manager object, or set the ignoreImport property in the policy.

Managing file references

An advantage of working with ontologies is that we can reuse work done by other ontologists, by importing their published ontologies into our own. Sometimes, however, this means that there is an Internet firewall between the ontology-based application and the source of an imported ontology. Even if it's possible to traverse the firewall through an HTTP proxy, retrieving files from an HTTP connection may impose unacceptable delays when starting an application. In the worst case, we may find that ontology on which our application depends is temporarily or permanently unavailable from the original published URL. To alleviate these commonly experienced problems, you can use the ontology document manager to set up a a local indirection, so that an attempt to import a document from a given published URL means that a local copy of the document is retrieved instead. This may be a file copy, or simply a pointer to a local mirror web site.

To specify this local redirection in the policy file, use the following declarations:

<OntologySpec>
  <publicURI rdf:resource="… the public URI to map from…"    />
  <altURL rdf:resource="… the local URL to map to …" />
  <!-- optional ontology language term -->
  <language rdf:resource="… encoding used …" />
  <!-- optional prefix to associate with the public URL -->
  <prefix rdf:datatype="&xsd;string">a prefix</prefix>
</OntologySpec>

For example:

<OntologySpec>
  <!-- local version of the RDFS vocabulary -->
  <publicURI rdf:resource="http://www.w3.org/2000/01/rdf-schema"    />
  <altURL rdf:resource="file:vocabularies/rdf-schema.rdf" />
</OntologySpec>

This specifies that an attempt to load the RDFS vocabulary from http://www.w3.org/2000/01/rdf-schema will transparently cause file:vocabularies/rdf-schema.rdf to be fetched instead. You can specify any number of such re-directions in the policy file, or you can add them to the document manager object directly by calling the various setter methods (see the Javadoc for details). As a side-effect, this mechanism also means that ontologies may be named with any legal URI (not necessarily resolvable) - so long as the altURL is itself resolvable.

Note that the OntDocumentManager is an application of Jena's File Manager. See the notes on FileManager for details of additional options.

In the following example, I programmatically declare that the ESWC ontology is replicated locally on disk, and then I load it using the public URL. Assume that the constant JENA has been initialised to the directory in which Jena was installed.

OntModel m = ModelFactory.createOntologyModel();
  OntDocumentManager dm = m.getDocumentManager();
  dm.addAltEntry( "http://www.eswc2006.org/technologies/ontology",
                  "file:" + JENA + "src-examples/data/eswc-2006-09-21.rdf"    );
  m.read( "http://www.eswc2006.org/technologies/ontology" );

Specifying prefixes

A model keeps a table of URI prefixes that can be used to render URI's in the shortened prefix:name form, which is useful in displaying URI's in a readable way in user interfaces, and is essential in producing legal XML names that denote arbitrary URI's. The ontology model's table of prefixes can be initialized from a table kept by the document manager, which contains the standard prefixes plus any that are declared by in the policy file (or added to subsequently by method calls). To prevent the model's prefix table from being initialized in this way, use the property useDeclaredNsPrefixes in the policy file (with value false), or call the method setUseDeclaredPrefixes on the ontology object.

Caching of imported models

You can use the document manager to assist with loading ontology documents through its cache. Suppose two ontologies, A and B both import ontology C. We would like not to have to read C twice when loading A and then B. The document manager supports this use case by optionally caching C's model, indexed by URI. When A tries to import C, there is no cached copy, so a new model is created for C, the contents of C's URL read in to the model, then the C model is used in the compound document for A. Subsequently, when ontology B is loading imports, the document manager checks in its cache and finds an existing copy of C. This will be used in preference to reading a fresh copy of C from C's source URL, saving both time and storage space.

Caching of import models is switched on by default. To turn it off, use the policy property cacheModels, or call the method setCacheModels( boolean caching ) with caching = false. The document manager's current model cache can be cleared at any time by calling clearCache().

The generic ontology type: OntResource

All of the classes in the ontology API that represent ontology values have OntResource as a common super-class. This makes OntResource a good place to put shared functionality for all such classes, and makes a handy common return value for general methods. The Java interface OntResource extends Jena's RDF Resource interface, so any general method that accepts a resource or an RDFNode will also accept an OntResource, and consequently, any other ontology value.

Some of the common attributes of ontology resources that are expressed through methods on OntResource are shown below:

For each of these properties, there is a standard pattern of available methods:

For example: addSameAs( Resource r ), or isSameAs( Resource r ). For full details of the individual methods, please consult the Javadoc.

OntResource defines some other general utility methods. For example, to find out how many values a resource has for a given property, you can call getCardinality( Property p ). To delete the resource from the ontology altogether, you can call remove(). The effect of this is to remove every statement that mentions this resource as a subject or object of a statement.

To get or set the value of a given property, use getPropertyValue( Property p ) or setPropertyValue( Property p, RDFNode value ). Continuing the naming pattern, the values of a named property can be listed (with listPropertyValues), removed (with removeProperty) or added (with addProperty).

Finally, OntResource provides methods for listing, getting and setting the rdf:type of a resource, which denotes a class to which the resource belongs (remember that in RDF and OWL, a resource can belong to many classes at once). The rdf:type property is one for which many entailment rules are defined in the semantic models of the various ontology languages. Therefore, the values that listRDFTypes() returns is more than usually dependent on the actual reasoner bound to the ontology model. For example, suppose I have class A, class B which is a subclass of A, and resource x whose asserted rdf:type is B. With no reasoner, listing x's RDF types will return only B. If the reasoner is able to calculate the closure of the subclass hierarchy (and most can), X's RDF types would also include A. A complete OWL reasoner would also infer that x has rdf:type owl:Thing and rdf:Resource.

For some tasks, getting a complete list of the RDF types of a resource is exactly what is needed. For other tasks, this is not the case. If you are developing an ontology editor, for example, you may want to distinguish in its display between inferred and asserted types. In the above example, only x rdf:type B is asserted, everything else is inferred. One way to make this distinction is to make use of the base model (see Figure 4). Getting the resource from the base model and listing the type properties there would return only the asserted values. For example:

// create the base model
String SOURCE = "http://www.eswc2006.org/technologies/ontology";
String NS = SOURCE + "#";
OntModel base = ModelFactory.createOntologyModel( OWL_MEM );
base.read( SOURCE, "RDF/XML" );

// create the reasoning model using the base
OntModel inf = ModelFactory.createOntologyModel( OWL_MEM_MICRO_RULE_INF, base );

// create a dummy paper for this example
OntClass paper = base.getOntClass( NS + "Paper" );
Individual p1 = base.createIndividual( NS + "paper1", paper );

// list the asserted types
for (Iterator i = p1.listRDFTypes(); i.hasNext(); ) {
    System.out.println( p1.getURI() + " is asserted in class " + i.next() );
}

// list the inferred types
p1 = inf.getIndividual( NS + "paper1" );
for (Iterator i = p1.listRDFTypes(); i.hasNext(); ) {
    System.out.println( p1.getURI() + " is inferred to be in class " + i.next() );
}

For other user interface or presentation tasks, we may want something between the complete list of types and the base list of only the asserted values. Consider the class hierarchy in figure 5 (i):

Figure 5: asserted and inferred relationships

Figure 5 (i) shows a base model, containing a class hierarchy and an instance x. Figure 5 (ii) shows the full set of relationships that might be inferred from this base model. In Figure 5 (iii), we see only the direct or maximally specific relationships. For example, in 5 (iii) x does not have rdf:type A, since this is an relationship that is covered by the facts that x has rdf:type D, and D is a subclass of A. Notice also that the rdf:type B link is also removed from the direct graph, for a similar reason. Thus the direct graph hides relationships from both the inferred and asserted graphs. When displaying instance x in a GUI, particularly in a tree view of some kind, the direct graph is often the most useful as it contains the useful information in the most compact form.

To list the RDF types of a resource, use:

listRDFTypes()                 // assumes not-direct
listRDFTypes( boolean direct ) // if direct=true, show only direct relationships

Related methods allow the rdf:type to be tested, set and returned.

Ontology classes and basic class expressions

Classes are the basic building blocks of an ontology. A simple class is represented in Jena by an OntClass object. As I mentioned above, an ontology class is a facet of an RDF resource. One way, therefore, to get an ontology class is to convert an RDF resource. Assume that m is a suitable defined OntModel, into which the ESWC ontology has already been read, and that NS is a variable denoting the ontology namespace:

Resource r = m.getResource( NS + "Paper" );
OntClass paper = (OntClass) r.as( OntClass.class );

This can be shortened by calling getOntClass() on the ontology model:

OntClass paper = m.getOntClass( NS + "Paper" );

The getOntClass method will retrieve the resource with the given URI, and attempt to obtain the OntClass facet. If either of these operations fail, getOntClass() will return null. Compare this with the createClass method, which will reuse an existing resource if possible, or create a new class resource if not:

OntClass paper     = m.createClass( NS + "Paper" );
OntClass bestPaper = m.createClass( NS + "BestPaper" );

You can use the create class method to create an anonymous class – a class description with no associated URI. Anonymous classes are often used when building more complex ontologies in OWL or DAML. They are less useful in RDFS.

OntClass anonClass = m.createClass();

Once you have the ontology class object, you can begin processing it through the methods defined on OntClass. The attributes of a class are handled in a similar way to the attributes of OntResource, above, with a collection of methods to set, add, get, test, list and remove values. Properties of classes that are handled in this way are:

Thus, in our example ontology, we can print a list the subclasses of an Artefact as follows:

OntClass artefact = m.getOntClass( NS + "Artefact" );
for (Iterator i = artefact.listSubClasses(); i.hasNext(); ) {
  OntClass c = (OntClass) i.next();
  System.out.println( c.getURI() );
}

Note that, under RDFS and OWL semantics, each class is a sub-class of itself (in other words, rdfs:subClassOf is reflexive). While this is true, under the semantics, Jena users have reported finding it unhelpful. Therefore, the listSubClasses and listSuperClasses methods remove the reflexive from the list of results returned by the iterator. However, if you use the plain Model API to query for rdfs:subClassOf triples, assuming that a reasoner is in use, the reflexive triple will appear in the deductions model.

Given an OntClass object, you can create or remove members of the class extension – individuals that are instances of the class – using the following methods:

To test whether a class is a root of the class hierarchy in this model (i.e. it has no known super-classes), call isHierarchyRoot().

The domain of a property is intended to allow entailments about the class of an individual, given that it appears as a statement subject. It is not a constraint that can be used to validate a document, in the way that XML schema can do. Nevertheless, many developers find it convenient to use the domain of a property to document the design intent that the property only applies to known instances of the domain class. Given this observation, it can be a useful debugging or display aide to show the properties that have this class among their domain classes. The method listDeclaredProperties() attempts to identify the properties that are intended to apply to instances of this class. Using listDeclaredProperties is explained in detail in the RDF frames how-to.

Ontology properties

In an ontology, a property denotes the name of a relationship between resources, or between a resource and a data value. It corresponds to a predicate in logic representations. One interesting aspect of languages like RDFS and OWL is that properties are not defined as aspects of some enclosing class, but are first-class objects in their own right. This means that ontologies and ontology-applications can store, retrieve and make assertions about properties directly. Consequently, Jena has a set of Java classes that allow you to conveniently manipulate the properties represented in an ontology model.

A property in an ontology model is an extension of the core Jena API class Property, and allows access to the additional information that can be asserted about properties in an ontology language. The common API super-class for representing ontology properties in Java is OntProperty. Again, using the pattern of add, set, get, list, has, and remove methods, we can access the following attributes of an OntProperty:

In the example ontology, the property hasProgramme has a domain of OrganizedEvent, a range of Programme and the human-readable label "has programme". I can reconstruct this definition in an empty ontology model as follows:

OntModel m = ModelFactory.createOntologyModel( OntModelSpec.OWL_MEM );
OntClass programme = m.createClass( NS + "Programme" );
OntClass orgEvent = m.createClass( NS + "OrganizedEvent" );

ObjectProperty hasProgramme = m.createObjectProperty( NS + "hasProgramme" );

hasProgramme.addDomain( orgEvent );
body.addRange( programme );
body.addLabel( "has programme", "en" );

As a further example, I can alternatively add information to an existing ontology. Let's add a super-property hasDeadline to generalise the separate properties denoting the submission deadline, notification deadline and camera-ready deadline:

OntModel m = ModelFactory.createOntologyModel( OntModelSpec.OWL_MEM );
m.read( "http://www.eswc2006.org/technologies/ontology" );

DatatypeProperty subDeadline = m.getDatatypeProperty( NS + "hasSubmissionDeadline" );
DatatypeProperty notifyDeadline = m.getDatatypeProperty( NS + "hasNotificationDeadline" );
DatatypeProperty cameraDeadline = m.getDatatypeProperty( NS + "hasCameraReadyDeadline" );

DatatypeProperty deadline = m.createDatatypeProperty( NS + "deadline" );
deadline.addDomain( m.getOntClass( NS + "Call" ) );
deadline.addRange( XSD.dateTime );

deadline.addSubProperty( subDeadline );
deadline.addSubProperty( notifyDeadline );
deadline.addSubProperty( cameraDeadline );

Note that, although I called the addSubProperty method on the object representing the new super-property, the serialized form of the ontology will contain rdfs:subPropertyOf axioms on each of the sub-property resources, since this is what the language defines. Jena will, in general, try to allow symmetric access to sub-properties and sub-classes from either direction.

Object and Datatype properties

OWL and DAML+OIL refine the basic property type from RDF into two sub-types: object properties and datatype properties (for more details see [OWL Reference]). The difference between them is that an object property can have only individuals in its range, while a datatype property has concrete data literals (only) in its range. Some OWL reasoners are able to exploit the differences between object and datatype properties to perform more efficient reasoning over ontologies. OWL also adds an annotation property, which is defined to have no semantic entailments, and so is useful when annotating ontology documents, for example.

In Jena, the Java interfaces ObjectProperty, DatatypeProperty and AnnotationProperty are sub-types of OntProperty. However, they do not have any behaviours (methods) particular to themselves. Their existence allows the more complex sub-types of ObjectProperty – transitive properties and so forth – to be kept separate in the class hierarchy. However, when you create an object property or datatype property in a model, it will have the effect of asserting different rdf:type statements into the underlying triple store.

Functional properties

OWL permits object and datatype properties to be functional – that is, for a given individual in the domain, the range value will always be the same. In particular, if father is a functional property, and individual :bryn has father :ijd and father :ian, a reasoner is entitled to conclude that :ijd and :ian denote the same individual. A functional property is equivalent to stating that the property has a maximum cardinality of one.

Being a functional property is represented through the FunctionalProperty facet of an ontology property object. If a property is declared functional (test using the isFunctional() method), then the method asFunctionalProperty() conveniently returns the functional facet. A non-functional property can be made functional through the convertToFunctionalProperty() method. When you are creating a property object, you also have the option of passing a Boolean parameter to the createObjectProperty() method on OntModel.

Other property types

There are several additional sub-types of ObjectProperty that represent additional capabilities of ontology properties. A TransitiveProperty means that if p is transitive, and we know :a p :b and also b p :c, we can infer that :a p :c. A SymmetricProperty means that if p is symmetric, and we know :a p :b, we can infer :b p :a. An InverseFunctionalProperty means that for any given range element, the domain value is unique. (Note that in DAML+OIL, the terminology for inverse functional property is unambiguous property).

Given that all properties are RDFNode objects, and therefore support the as() method, you can use as() to change from an object property facet to a transitive property facet. To make this more straightforward, the OntProperty Java class has a number of methods that support directly switching to the corresponding facet view:

public TransitiveProperty asTransitiveProperty();
public FunctionalProperty asFunctionalProperty();
public SymmetricProperty asSymmetricPropery();
public InverseFunctionalProperty asInverseFunctionalProperty();

These methods all assume that the underlying model will support this change in perspective. If not, the operation will fail with a ConversionException. For example, if a given property p is not already a transitive property, then invoking p.asTransitiveProperty() will throw a conversion exception. The following methods will, if necessary, add additional information (i.e. the additional rdf:type statement) to allow the conversion to an alternative facet to succeed.

public TransitiveProperty convertToTransitiveProperty();
public FunctionalProperty convertToFunctionalProperty();
public SymmetricProperty convertToSymmetricPropery();
public InverseFunctionalProperty convertToInverseFunctionalProperty();

Finally, methods beginning is… (e.g. isTransitiveProperty) allow you to test whether a given property would support a given sub-type facet.

More complex class expressions

I introduced the handling of basic, named classes above. These are the only kind of class descriptions available in RDFS. In OWL, however, there are a number of additional types of class expression, which allow richer and more expressive descriptions of concepts. There are two main categories of additional class expression: restrictions and Boolean expressions. Let's examine each in turn.

Restriction class expressions

A restriction defines a class by reference to one of the properties of the individuals that comprise the members of the class, and then placing some constraint on that property. For example, in a simple view of animal taxonomy, we might say that mammals are covered in fur, and birds in feathers. Thus the property hasCovering is in one case restricted to have the value fur, in the other to have the value feathers. This is a has value restriction. Six restriction types are currently defined by OWL:

Note that in DAML+OIL terminology, an all-values-from restriction is a toClass restriction, while a some-values-from restriction is a hasClass restriction. Note also that, at present, the Jena ontology API has only limited support for DAML's qualified cardinality restrictions (i.e. cardinalityQ, minCardinalityQ and maxCardinalityQ). Qualified cardinality restrictions are encapsulated in the interfaces CardinalityQRestriction, MinCardinalityQRestriction and CardinalityQRestriction. OntModel also provides methods for creating and accessing qualified cardinality restrictions. Since they are not part of the OWL 1.0 language definition, qualified cardinality restrictions are not supported in OWL ontologies. Qualified cardinality restrictions may be added to the forthcoming OWL 1.1 update, at which point it is likely that the Jena ontology API will be updated. Note however, that we are not currently making any commitment on the timescale for such an update.

Jena provides a number of ways of creating restrictions, or retrieving them from a model. Firstly, you can retrieve a general restriction from the model by its URI, if known.

// get restriction with a given URI
Restriction r = m.getRestriction( NS + "theName" );

You can create a new restriction created by nominating the property that the restriction applies to:

// anonymous restriction on property p
OntProperty p = m.createOntProperty( NS + "p" );
Restriction anonR = m.createRestriction( p );

Since a restriction is typically not assigned a URI in an ontology, retrieving an existing restriction by name may not be possible. However, you can list all of the restrictions in a model and search for the one you want:

Iterator i = m.listRestrictions();
while (i.hasNext()) {
    Restriction r = (Restriction) i.next();
    if (isTheOne( r )) {
        // handle the restriction
    }
}

A common case is that you want the restrictions on some property p. In this case, from an object denoting p you can list the restrictions that mention that property:

OntProperty p = m.getProperty( NS + "p" );
Iterator i = p.listReferringRestrictions();
while (i.hasNext()) {
    Restriction r = (Restriction) i.next();
    // now handle the restriction …
}

A general restriction can be converted to a specific type of restriction via as… methods (if the information is already in the model), or, if the information is not in the model, via convertTo… methods. For example, to convert the example restriction r from the example above to an all values from restriction, you can do the following:

OntClass c = m.createClass( NS + "SomeClass" );
AllValuesFromRestriction avf = r.convertToAllValuesFromRestriction( c );

To create a particular restriction ab initio, you can use the creation methods defined on OntModel. For example:

OntClass c = m.createClass( NS + "SomeClass" );
ObjectProperty p = m.createObjectProperty( NS + "p" );

// null denotes the URI in an anonymous restriction
AllValuesFromRestriction avf = m.createAllValuesFromRestriction( null, p, c );

Assuming that the above code fragment was using a model m that was created with the OWL language profile, it creates a instance of an OWL restriction that would have the following definition in RDF/XML:

<owl:Restriction>
  <owl:onProperty rdf:resource="#p"/>
  <owl:allValuesFrom rdf:resource="#SomeClass"/>
</owl:Restriction>

Once you have a particular restriction object, there are methods following the standard add, get, set and test naming pattern to access the aspects of the restriction. For example, in a camera ontology, we might find this definition of a class describing Large-Format cameras:

<owl:Class rdf:ID="Large-Format">
  <rdfs:subClassOf rdf:resource="#Camera"/>
  <rdfs:subClassOf>
    <owl:Restriction>
      <owl:onProperty rdf:resource="#body"/>
      <owl:allValuesFrom rdf:resource="#BodyWithNonAdjustableShutterSpeed"/>
   </owl:Restriction>
  </rdfs:subClassOf>
</owl:Class>

Here's one way to access the components of the all values from restriction. Assume m contains a suitable camera ontology:

OntClass largeFormat = m.getOntClass( camNS + "Large-Format" );
for (Iterator i = LargeFormat.listSuperClasses( true ); i.hasNext(); ) {
  OntClass c = (OntClass) i.next();

  if (c.isRestriction()) {
    Restriction r = c.asRestriction();

    if (r.isAllValuesFromRestriction()) {
      AllValuesFromRestriction av = r.asAllValuesFromRestriction();
      System.out.println( "AllValuesFrom class " +
                          av.getAllValuesFrom().getURI() +
                          " on property " + av.getOnProperty().getURI() );
    }
  }
}

Boolean class expressions

Most programmers are familiar with the use of Boolean operators to construct propositional expressions: conjunction (and), disjunction (or) and negation (not). OWL provides a means for construction expressions describing classes with analogous operators, by considering class descriptions in terms of the set of individuals that comprise the members of the class.

Suppose we wish to say that an instance x has rdf:type A and rdf:type B. This means that x is both a member of the set of individuals in A, and in the set of individuals in B. Thus, x lies in the intersection of classes A and B. If, on the other hand, A is either has rdf:type A or B, then x must lie in the union of A and B. Finally, to say that x does not have rdf:type A, it must lie in the complement of A. These operations, union, intersection and complement are the Boolean operators for constructing class expressions. While complement takes only a single argument, union and intersection must necessarily take more than one argument. Before continuing with constructing and using Boolean class expressions, I digress briefly to discuss lists.

List expressions

RDF originally had three container types: Seq, Alt and Bag. While useful, these are all open forms: it is not possible to say that a given container has a fixed number of values. The DAML+OIL standard introduced a fourth container type - lists - in order to have a closed collection. Lists have subsequently been added to the core RDF specification, and are used extensively in OWL. A list follows the well-known cons cell pattern from Lisp, Prolog and other list-handling languages. Each cell of a list is either the end-of-list terminator (nil in Lisp), or is a pair consisting of a value and a pointer to the cell that is the first cell on the tail of the list. In RDF lists, the end-of-list is marked by a resource with name rdf:nil, while each list cell is an anonymous resource with two properties, one denoting the tail and the other the value. Fortunately, this complexity is hidden by some simple syntax:

<p rdf:parseType="collection">
  <A />
  <B />
</p>

According to the RDF specification, this list of two elements has the following expansion as RDF triples:

<p>
  <rdf:first><A /></rdf:first>
  <rdf:rest>
    <rdf:first><B /></rdf:first>
    <rdf:rest rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#nil"/>
  </rdf:rest>
</p>

Given this construction, a well formed list (one with exactly one rdf:first and rdf:rest per cons cell) has a precisely determined set of members. Incidentally, the same list in N3/Turtle is even more compact:

:example
    :p ( :A :B ).

Although lists are defined in the generic RDF model in Jena, they are extensively used by the ontology API so I mention them here. Full details of the methods defined on the RDFList class are available in the Jena Javadoc.

A number of means of constructing lists is defined in Model, as variants on createList. For example, you can construct a list of three classes as follows:

OntModel m = ModelFactory.createOntModel();
OntClass c0 = m.createClass( NS + "c0" );
OntClass c1 = m.createClass( NS + "c1" );
OntClass c2 = m.createClass( NS + "c2" );

RDFList cs = m.createList( new RDFNode[] {c0, c1, c2} );

Alternatively, you can build a list one element at at time:

OntModel m = ModelFactory.createOntModel();
RDFList cs = m.createList(); // Cs is empty
cs = cs.cons( m.createClass( NS + "c0" ) );
cs = cs.cons( m.createClass( NS + "c1" ) );
cs = cs.cons( m.createClass( NS + "c2" ) );

Note that these two approaches end with the classes in the lists in opposite orders, since the cons operation adds a new list cell to the front of the list. Thus the second list will run c2 to c0. In the ontology operations we are discussing here, the order of values in the list is not considered significant.

Once the list has been constructed or obtained from the model (a resource which is a cell in a list sequence will accept .as( RDFList.class )), RDFList methods may be used to access members of the list, iterate over the list, and so forth. For example:

System.out.println( "List has " + myRDFList.size() + " members:" );
for (Iterator i = myRDFList.iterator(); i.hasNext(); ) {
  System.out.println( i.next() );
}

Intersection, union and complement class expressions

Given Jena's ability to construct lists, building intersection and union class expressions is straightforward. The create methods on OntModel allow you to construct an intersection or union directly. Alternatively, given an existing OntClass, you can use the convertTo… methods to construct facet representing the more specialised expressions. For example, we can define the class of UK industry-related conferences as the intersection of conferences with a UK location and conferences with an industrial track. Here's the XML declaration:

<owl:Class rdf:ID="UKIndustrialConference">
  <owl:intersectionOf rdf:parseType="Collection">
    <owl:Restriction>
      <owl:onProperty rdf:resource="#hasLocation"/>
      <owl:hasValue rdf:resource="#united_kingdom"/>
    </owl:Restriction>
  </owl:intersectionOf>
    <owl:Restriction>
      <owl:onProperty rdf:resource="#hasPart"/>
      <owl:someValuesFrom rdf:resource="#IndustryTrack"/>
    </owl:Restriction>
  </owl:intersectionOf>
</owl:Class>

Or, more compactly in N3/Turtle:

:UKIndustrialConference a owl:Class
    ; owl:intersectionOf (
       [a owl:Restriction
        ; owl:onProperty :hasLocation
        ; owl:hasValue :united_kingdom]
       [a owl:Restriction
        ; owl:onProperty :hasPart
        ; owl:someValuesFrom :IndustryTrack]
      )

Here is code to create this class declaration using Jena, assuming that m is a model into which the ESWC ontology has been read:

// get the class references
OntClass place = m.getOntClass( NS + "Place" );
OntClass indTrack = m.getOntClass( NS + "IndustryTrack" );

// get the property references
ObjectProperty hasPart = m.getObjectProperty( NS + "hasPart" );
ObjectProperty hasLoc = m.getObjectProperty( NS + "hasLocation" );

// create the UK instance
Individual uk = place.createIndividual( NS + "united_kingdom" );

// now the anonymous restrictions
HasValueRestriction ukLocation =
    m.createHasValueRestriction( null, hasLoc, uk );
SomeValuesFromRestriction hasIndTrack =
    m.createHasValueRestriction( null, hasPart, indTrack );

// finally create the intersection class
IntersectionClass ukIndustrialConf =
    m.createIntersectionClass( NS + "UKIndustrialConference",
                               m.createList( new RDFNode[] {ukLocation, hasIndTrack} ) );

Given the similarity between union and intersection class expressions, rather than separate methods to set the components of the expression, Jena defines a common super-class BooleanClassDescription. This class provides access to the operands to the expression. In the above example, the operands are the two restrictions. The BooleanClassDescription class allows you to set the operands en masse by supplying a list, or to be added or deleted one at a time.

Complement class expressions are very similar. The principal difference is that they take only a single class as operand, and therefore do not accept a list of operands.

Enumerated classes

The final type class expression allows by OWL is the enumerated class. Recall that a class is a set of individuals. Often, we want to define the members of the implicitly: for example, "the class of UK conferences". Sometimes it is convenient to define a class explicitly, by stating the individuals the class contains. An enumerated class is exactly the class whose members are the given individuals. For example, we know that the class of PrimaryColours contains exactly red, green and blue, and no others.

In Jena, an enumerated class is created in a similar way to other classes. The set of values that comprise the enumeration is described by an RDFList. For example, here's a class defining the countries that comprise the United Kingdom:

<owl:Class rdf:ID="UKCountries">
  <owl:oneOf rdf:parseType="Collection">
    <eswc:Place rdf:about="#england"/>
    <eswc:Place rdf:about="#scotland"/>
    <eswc:Place rdf:about="#wales"/>
    <eswc:Place rdf:about="#northern_ireland"/>
  </owl:oneOf>
</owl:Class>

To list the contents of this enumeration, we could do the following:

OntClass place = m.getOntClass( NS + "Place" );

EnumeratedClass ukCountries =
    m.createEnumeratedClass( NS + "UKCountries", null );
ukCountries.addOneOf( place.createIndividual( NS + "england" ) );
ukCountries.addOneOf( place.createIndividual( NS + "scotland" ) );
ukCountries.addOneOf( place.createIndividual( NS + "wales" ) );
ukCountries.addOneOf( place.createIndividual( NS + "northern_ireland" ) );

for (Iterator i = UKCountries.listOneOf(); i.hasNext(); ) {
  Resource r = (Resource) i.next();
  System.out.println( r.getURI() );
}

Listing classes

In many applications, you will need to inspect the set of classes in an ontology. The list… methods on OntModel provide a variety of means of listing types of class. The methods available include:

public ExtendedIterator listClasses();
public ExtendedIterator listEnumeratedClasses();
public ExtendedIterator listUnionClasses();
public ExtendedIterator listComplementClasses();
public ExtendedIterator listIntersectionClasses();
public ExtendedIterator listRestrictions();
public ExtendedIterator listNamedClasses();
public ExtendedIterator listHierarchyRootClasses();

The last two methods deserve special mention. In OWL, class expressions are typically not named, but are denoted by anonymous resources (aka bNodes). In many applications, such as displaying an ontology in a user interface, we want to pick out the named classes only, ignoring those denoted by bNodes. This is what listNamedClasses() does. The method listHierarchyRootClasses() identifies the classes that are uppermost in the class hierarchy contained in the given model. These are the classes that have no super-classes. The iteration returned by listHierarchyRootClasses() may contain anonymous classes. To get a list of named hierarchy root classes, i.e. the named classes that lie closest to the top of the hierarchy (alternatively: the shallowest fringe of the hierarchy consisting solely of named classes), use the OntTools method namedHierarchyRoots().

You should also note that it is important to close the iterators returned from the list… methods, particularly when the underlying store is a database. This is necessary so that any state (e.g. the database connection resources) can be released. Closing happens automatically when the hasNext() method on the iterator returns false. If your code does not iterate all the way to the end of the iterator, you should call the close() method explicitly. Note also that the values returned by these iterators will depend on the asserted data and the reasoner being used. For example, if the model contains a Restriction, that restriction will only be returned by the listClasses() iterator if the model is bound to a reasoner that can infer that any restriction is also be a class, since Restriction is a subClassOf Class. This difference can be exploited by the programmer: to list classes and restrictions separately, perform the listClasses() and listRestictions() methods on the base model only, or on a model with no reasoner attached.

Instances or individuals

In OWL Full (and in DAML+OIL) any value can be an individual – and thus the subject of triples in the RDF graph other than ontology declarations. In OWL Lite and DL, the language terms and the instance data that the application is working with are kept separate, by definition of the language. Jena therefore supports a simple notion of an Individual, which is essentially an alias for Resource. While Individuals are largely synonymous with Resources, they do provide an programming interface that is consistent with the other Java classes in the ontology API.

There are two ways to create individuals. Both requires the class to which the individual will initially belong:

OntClass c = m.createClass( NS + "SomeClass" );

// first way: use a call on OntModel
Individual ind0 = m.createIndividual( NS + "ind0", c );

// second way: use a call on OntClass
Individual ind1 = c.createIndividual( NS + "ind1" );

The only real difference between these approaches is that the second way will create the individual in the same model that the class is attached to (see the getModel() method). In both of the above examples the individual is named, but this is not necessary. The method OntModel.createIndividual( Resource cls ) creates an anonymous individual belonging to the given class. Note that the type of the class parameter is only Resource. You are not required to use as() to present a Resource to an OntClass before calling this method, though of course an OntClass is a Resource so using an OntClass will work perfectly well.

Individual provides a set of methods for testing and manipulating the ontology classes to which an individual belongs. This is a convenience: OWL and RDFS denote class membership through the rdf:type property, and methods for manipulating and testing rdf:type are defined on OntResource. You may use either approach interchangeably.

Ontology meta-data

In OWL and DAML+OIL, but not RDFS, meta-data about the ontology itself is encoded as properties on an individual of class owl:Ontology or daml:Ontology as appropriate. By convention, the URI of this individual is the URI of the ontology document itself. In the XML serialisation, this is typically shown as:

<owl:Ontology rdf:about="">
</owl:Ontology>

Note that the construct rdf:about="" does not indicate a resource with no URI; it is in fact a shorthand way of referencing the base URI of the document containing the ontology. The base URI may be stated in the document through an xml:base declaration in the XML preamble. The base URI can also be specified when reading the document via Jena's Model API (see the read() methods on OntModel for reference).

You can attach various meta-data statements to this object to indicate attributes of the ontology as a whole. The Java object Ontology represents this special instance, and uses the standard add, set, get, list, test and delete pattern to provide access to the following attributes:

In addition to these attributes, the Ontology element typically contains common meta-data properties, such as comment, label and version information.

In the Jena API, the ontology's metadata properties can be accessed through the Ontology interface. Suppose I wish to know the list of URI's that the ontology imports. First I must obtain the resource representing the ontology itself:

String base = …; // the base URI of the ontology
OntModel m = …;  // the model containing the ontology statements
Ontology ont = m.getOntology( base );

// now list the ontology imports
for (Iterator i = ont.listImports(); i.hasNext(); ) {
    System.out.println( "Ontology " + base + " imports " + i.next() );
}

If the base URI of the ontology is not known, you can list all resources of rdf:type Ontology in a given model by OntModel.listOntologies(). If there is only one of these, it should be safe to assume that it is the Ontology resource for the ontology. However, you should note that if more than one ontology document has been read in to the model (for example by including the imports of a document), there may well be more than one Ontology resource in the model. In this case, you may find it useful to list the ontology resources in just the base model:

OntModel m = … // the model, including imports
OntModel mBase = ModelFactory.createOntologyModel(
                      OntModelSpec.OWL_MEM, m.getBaseModel() );

for (Iterator i = mBase.listOntologies(); i.hasNext(); ) {
    Ontology ont = (Ontology) i.next();
    // m's base model has ont as an import ...
}

A common practice is also to use the Ontology element to attach Dublin Core metadata to the ontology document. Jena provides a copy of the Dublin Core vocabulary, in com.hp.hpl.jena.vocabulary.DC. To attach a statement saying that the ontology was authored by John Smith, we can say:

Ontology ont = m.getOntology( baseURI );
ont.addProperty( DC.creator, "John Smith" );

It is also possible to programatically add imports and other meta-data to a model, for example:

String base = …; // the base URI of the ontology
OntModel m = …;

Ontology ont = m.createOntology( base );
ont.addImport( m.createResource( "http://example.com/import1" ) );
ont.addImport( m.createResource( "http://example.com/import2" ) );

Note that under default conditions, simply adding (or removing) an owl:imports statement to a model will not cause the corresponding document to be imported (or removed). However, by calling OntModel.setDynamicImports(true), the model will start noticing the addition or removal of owl:imports statements.

Ontology inference: overview

You have the choice of whether to use the Ontology API with Jena's reasoning capability turned on, and, if so, which of the various reasoners to use. Sometimes a reasoner will add information to the ontology model that it is not useful for your application to see. A good example is an ontology editor. Here, you may wish to present your users with the information they have entered in to their ontology; the addition of the entailed information into the editor's display would be very confusing. Since Jena does not have a means for distinguishing inferred statements from those statements asserted into the base model, a common choice for ontology editors and similar applications is to run with no reasoner.

In many other cases, however, it is the addition of the reasoner that makes the ontology useful. For example, if we know that John is the father of Mary, we would expect a 'yes' if we query whether John is the parent of Mary. The parent relationship is not asserted, but we know from our ontology that fatherOf is a sub-property of parentOf. If 'John fatherOf Mary' is true, then 'John parentOf Mary' is also true. The integrated reasoning capability in Jena exists to allow just such entailments to be seen and used.

For a complete and thorough description of Jena's inference capabilities, please see the reasoner documentation. This section of of the ontology API documentation is intended to serve as only a brief guide and overview.

Recall from the introduction that the reasoners in Jena operate by making it appear that triples entailed by the inference algorithm are part of the model in just the same way as the asserted triples (see Figure 2). The underlying architecture allows the reasoner to be part of the same Java virtual machine (as is the case with the built-in rule-based reasoners), or in a separate process on the local computer, or even a remote machine. Of course, each of these choices will have different characteristics of what reasoning capabilities are supported, and what the implications for performance are.

The reasoner attached to an ontology model, if any, is specified through the OntModelSpec. The methods setReasoner() and setReasonerFactory() on the model spec are used to specify a reasoner. The setReasoner variant is intended for use on a specification which will only be used to build a single model. The factory variant is used where the OntModelSpec will be used to build more than one model, ensuring that each model gets its own reasoner object. The ReasonerRegistry provides a collection of pre-built reasoners – see the reasoner documentation for more details. However, it is also possible for you to define your own reasoner that conforms to the appropriate interface. For example, there is an in-process interface to the open-source Pellet reasoner.

To facilitate the choice of reasoners for a given model, some common choices have been included in the pre-built ontology model specifications available as static fields on OntModelSpec. The available choices are described in the section on ont model specifications, above.

Depending on which of these choices is made, the statements returned from queries to a given ontology model may vary considerably.

Additional notes

Jena's inference machinery defines some specialised services that are not exposed through the addition of extra triples to the model. These are exposed by the InfModel interface; for convenience OntModel extends this interface to make these services directly available to the user. Please note that calling inference-specific methods on an ontology model that does not contain a reasoner will have unpredictable results. Typically these methods will have no effect or return null, but you should not rely on this behaviour.

In general, inference models will add a lot of additional statements to a given model, including the axioms appropriate to the ontology language. This is typically not something you will want in the output when the model is serialized, so write() on an ontology model will only write the statements from the base model. This is typically the desired behaviour, but there are occasions (e.g. during debugging) when you may want to write the entire model, virtual triples included. The easiest way to achieve this is to call the writeAll() method on OntModel. An alternative technique, which can sometimes be useful for a variety of use-cases, including debugging, is to snapshot the model by constructing a temporary plain model and adding to it: the contents of the ontology model:

OntModel m = …

// snapshot the contents of ont model om
Model snapshot = ModelFactory.createDefaultModel();
snapshot.add( om );

See also the note on combining a-box and t-box data using the OntModel.

Working with DAML+OIL ontologies

Jena 2.5 update With effect from release 2.5 of Jena, the legacy DAML+OIL API was deprecated, and has been removed from the 2.6 release. The effect of this is that users of DAML+OIL using the legacy API (i.e. the classes in com.hp.hpl.jena.ontology.daml) should move their projects either to work with OWL, or at least to work with the DAML+OIL profile of the standard OntModel. The DAML+OIL OntModel profile will continue to be supported.

Generally speaking, we recommend that developers who want to work on Semantic Web applications should use RDFS or OWL, as their ontology language. OWL has had the benefit of a more rigorous design and significant input to its design from researchers with insights into the design of efficient reasoning algorithms. Nevertheless, there remains some legacy ontology data on the Semantic Web today encoded in DAML+OIL, the precursor to OWL. Jena 2 currently supports ontology processing in DAML+OIL as well as OWL and RDFS. However, DAML reasoning using the built-in Jena reasoners is more restricted than is the case with OWL.

Since Jena 2.6, the ontology API (as described in this document) is the only available route for processing DAML data. Using the language profile, the ontology API can handle RDFS, OWL and DAML sources with equal ease. Thus, OntClass can be used to represent a DAML class, an OWL class or an RDFS class.

Working with persistent ontologies

A common way to work with ontology data is to load the ontology axioms and instances at run-time from a set of source documents. This is a very flexible approach, but has limitations. In particular, your application must parse the source documents each time it is run. For large ontologies, this can be a source of significant overhead. Jena provides an implementation of the RDF model interface that stores the triples persistently in a database. This saves the overhead of loading the model each time, and means that you can store RDF models significantly larger than the computer's main memory, but at the expense of a higher overhead (a database interaction) to retrieve and update RDF data from the model. In this section I show how to use the ontology API with Jena's persistent database models.

For information on setting-up and accessing the database models themselves, see the DB documentation. I do not cover this aspect in this section. For a method of using a persistent database to manage ontology imports, see this article.

There are two somewhat separate requirements for persistently storing ontology data. The first is making the main or base model itself persistent. The second is re-using or creating persistent models for the imports of an ontology. These two requirements are handled slightly differently. A simple, but complete, example of using persistent database models for storing ontologies is included with Jena.

To retrieve a Jena model from the database API, we have to know its name. Fortunately, common practice for ontologies on the Semantic Web is that each is named with a URI. We use this URI to name the model that is stored in the database. Note carefully what is actually happening here: we are exploiting a feature of the database sub-system to make persistently stored ontologies easy to retrieve, but we are not in any sense resolving the URI of the model. Once placed into the database, the name of the model is treated as an opaque string.

To create a persistent model for the ontology http://example.org/Customers, we create a model maker that will access our underlying database, and use the ontology URI as the database name. We then take the resulting persistent model, and use it as the base model when constructing an ontology model:

Model base = getMaker().createModel( "http://example.org/Customers" );
OntModel m = ModelFactory.createOntologyModel( OntModelSpec.OWL_MEM_RULE_INF, base );

Here we assume that the getMaker() method returns a suitably initializedModelMaker that will open the connection to the database. This step only creates a persistent model named with the ontology URI. To initialise the content, we must either add statements to the model using the OntModel API, or do a one-time read from a document:

m.read( "http://example.org/Customers" );

Once this step is completed, the model contents may be accessed in future without needing to read again.

If the Customers ontology imports other ontologies, using owl:imports or daml:imports, the Jena Ontology API will build a union model containing the closure of the imports. Even if the base model is persistent, the predefined OntModelSpec objects only specify memory models to contain the imported ontologies, since memory models do not require any additional parameters.

To specify that the imported models should stored in, and retrieved from, the database, we must update the ontology spec object to use the model maker that encapsulates the database connection:

OntModelSpec spec = new OntModelSpec( OntModelSpec.OWL_MEM_RULE_INF );
// set the model maker for the base model
spec.setBaseModelMaker( getMaker() );
// set the model maker for imports
spec.setImportModelMaker( getMaker() );

This new model maker will then be used to generate persistent models named with the URI of the imported ontology, if it passed instead of OntModelSpec.OWL_MEM_RULE_INF to the createOntologyModel method of the model factory. Note that once the import has been loaded once into the database, it can be re-used by other ontologies that import it. Thus a given database will only contain at most one copy of each imported ontology.

Note on performance The built-in Jena reasoners, including the rule reasoners, make many small queries into the model in order to propagate the effects of rules firing. When using a persistent database model, each of these small queries creates an SQL interaction with the database engine. This is a very inefficient way to interact with a database system, and performance suffers as a result. Efficient reasoning over large, persistent databases is currently an open research challenge. Our best suggested work-around is, where possible, to snapshot the contents of the database-backed model into RAM for the duration of processing by the inference engine. An alternative solution, that may be applicable if your application does not write to the datastore often, is to precompute the inference closure of the ontology and data in-memory, then store that into a database model to be queried by the run-time application. Such an off-line processing architecture will clearly not be applicable to every application problem.

A sample program shows the above steps combined, to create an ontology in which both base model and imports are stored in a persistent database.

Experimental ontology tools

Starting with Jena release 2.6, the OntTools class provides a small collection of commonly-requested utilities for assisting with ontology processing. Given that this is a new feature, you should regard it as an experimental facility for the time being. We welcome feedback. The capabilities in OntTools are implemented as static methods. Currently available tools are:

• OntClass getLCA( OntModel m, OntClass u, OntClass v )
  Determine the lowest common ancestor for classes u and v. This is the class that is lowest in the class hierarchy, and which includes both u and v among its sub-classes.
• Path findShortestPath( Model m, Resource start, RDFNode end, Filter onPath )
  Breadth-first search, including a cycle check, to locate the shortest path from start to end, in which every triple on the path returns true to the onPath predicate.
• List namedHierarchyRoots( OntModel m )
  Compute a list containing the uppermost fringe of the class hierarchy in the given model which consists only of named classes.

Author: Ian Dickinson
Last modified: $Id: index.html,v 1.32 2009-02-24 15:15:20 ian_dickinson Exp $