The primary goal of the Semantic Web is to use URIs as a universal space to name anything, expanding from using URIs for webpages to URIs for \real objects and imaginary concepts," as phrased by Berners-Lee. This distinction has often been tied to the distinction between information resources, like webpages and multimedia les, and non-information resources, which are everything from real people to abstract concepts like `the integers.' Furthermore, the W3C has recommended not to use the same URI for information resources and non-information resources, and several communities like the Linked Data initiative are deploying this principle. The de nition put forward by the W3C, that information resources are things whose \essential nature is information" is a di cult distinction at best. For example, would the text of Moby Dick be an information resource? While this problem could safely be ignored up until recently, with the rise of Linked Data and projects like OKKAM, it appears that this problem should be modelled formally. An ontology called IRW (Identity and Reference on the Web) of various types of resources and their relationships, both for the hypertext Web and Linked Data, is presented. It builds upon Information Object Lite (an extension of DOLCE Ultra Lite for describing information objects) and IRE (an earlier ontology of and aligns with other work in this area. This ontology can be used as a tool to make Linked Data more self-describing and to allow inference to be used to test for membership in various classes of resources.
H.3.d [Information Technology and Systems]: Metadata
The key feature of the Semantic Web is not its use of knowledge representation technologies like ontologies and inference per se, but the introduction of these technologies to operate over Web resources as de ned by URIs. Early Semantic Web e orts forgot this, and treated URIs as just odd sorts of symbols. The Linked Data Tutorial provided a way for putting Semantic Web technologies in harmony with Web architecture, and now Linked Data is experiencing amazing growth. Yet, there is still debate within Web architecture circles as to what the de nition of a `information resource' is, a term crucial to Linked Data, and how terms like this relate to the pre-Semantic Web hypertext Web. We model the terms used in Linked Data and Web architecture using a lightweight formal ontology in OWL-DL, which we call IRW, for `Identity of Resources on the Web.' The hope is this ontology will clarify these debates and allow further development of a provenance-aware and semantically veri ed Linked Data Web.
Before trying to gure out the di erence between a `noninformation' and `information' resource, what is a resource? The W3C TAG state in their Architecture of the Web that `resource' is used in a general sense for whatever might be identi ed by a URI [?]. Previously, a resource was thought of as strictly to be for network-accessible objects such as webpages, since the term `resource' is de ned by Fielding in the rst HTTP RFC as \a network data object or service, identi ed by a URI". However, Berners-Lee broadened the concept of resource in his RFC 2396, stating that \a resource can be anything that has identity. Familiar examples include an electronic document, an image, a service (e.g., `today's weather report for Los Angeles'), and a collection of other resources. Not all resources are network `retrievable'; e.g., human beings, corporations, and bound books in a library can also be considered resources" [?].
One distinction that has been upheld by Hayes and others is the distinction between reference and access [?]. Making an analogy between URIs and names, access means \that the name provides a causal pathway to the thing, perhaps mediated by the Web" while reference means that \the name is being used to mention the thing," which may or may not coincide with access [?]. Something is then `Web-accessible' if it can accessed via the use of HTTP. This use of the term `resource' for both referring to non-Web accessible things and for naming Web-accessible things is continued in URI RFC 3986, the current IETF RFC, which states that \this speci cation does not limit the scope of what might be a resource; rather, the term `resource'... likewise, abstract concepts can be resources, such as the operators and operands of a mathematical equation, the types of a relationship (e.g., `parent' or `employee'), or numeric values (e.g., zero, one, and in nity)" [?]. It is precisely this ability to name things with URIs that aren't Web-accessible that de nes both the Semantic Web and Linked Data. However, unlike traditional Semantic Web applications, Linked Data allows Web-accessible associated descriptions, in both machine and human-readable forms, to be accessed from a URI for a non-information resource. The most obvious distinction is between a resource that could in principle be Web-accessible, like a webpage, and a resource that is not in principle Web-accessible, like the Ei el Tower itself. This distinction is given by the W3C TAG as the distinction between an information resource and something that may not be an information resource [?]. The W3C TAG then de ne an information resource as something \whose essential characteristics can be conveyed in a message," which is a controversial de nition [?]. As noted by the Linked Data tutorial, this implies there is another kind of resource, non-information resources, for things that are not possibly Web-accessible, like a URI whose primary purpose is to refer to the Ei el Tower [?]. Furthermore, one can distinguish `Web resources' (a subset of information resources) that are usually Web-accessible, such as web-pages, from things that simply carry information, like the text of Moby Dick, regardless of whether it is on the Web or not. Again, let us emphasize that some nd these distinctions very intuitive, while others do not. Lastly, in order to distinguish URIs for non-accessible things on the Semantic Web (the `Cool URIs for the Semantic Web') from the normal use of URIs on the hypertext Web, we call the former Semantic Web URIs [?]. In Web architecture circles, what are typically called `webpages' are just one kind of a `representations' of a resource [?]. In order to distinguish the use of the word `representation' in Web architecture circles from its normal usage, the word Web Representation is used in this paper to designate a more encompassing notion of representation of a resource, i.e. any set of bits that is `coming down the wire' in response to the use of the Web.
from URIs for non-information resources by use of redirection. This was codi ed by the W3C TAG when it o cially resolved httpRange-14 by saying that the 303 See Other HTTP header can serve to disambiguate between information resources and possible non-information resources. The o cial resolution by the TAG is given below as [?]: a URI, since the Ei el Tower itself is not an information resource, no Web representations are directly available. Instead, the agent gets a 303 See Other that in turn redirects them to an information resource that hosts Web representations about the Ei el Tower, such as http://dbpedia.org/page/Eiffel_Tower. When this URI returns the 200 status code in response to an HTTP GET request, the agent can infer that http://dbpedia.org/page/Eiffel_Tower/ is actually an information resource. The Semantic Web URI used to refer to the Ei el Tower itself, http://dbpedia.org/resource/Eiffel_Tower, could be any kind of resource and so could be a non-information resource [?]. This example is illustrated in Figure ??, using terms from the IRW ontology introduced in Section ??. An alternative to the 303 redirection is the hash convention, in which one uses the fragment identi er of a URI to get redirection `for free' with smaller RDF vocabularies. If one wanted a Semantic Web URI that referred to the Ei el Tower itself without the hassle of a 303 redirection, one would use the URI http://www.tour-eiffel.fr/#it to refer to the Ei el Tower itself. Since browsers either dispose of or treat the fragment identi er as a fragment of a hypertext document or some other Web representation, if an agent tries to access via HTTP GET a Semantic Web URI that uses the hash convention, the server will not return a 404 Not Found status code, but instead will resolve to the URI before the hash, http://www.tour-eiffel.fr, which can then be an Web resource capable of returning Web representations, which is called an `associated description' in the Linked Data community [?]. In this way, Semantic Web inference engines can keep the Semantic Web URI that refers to the Ei el Tower and an associated description about the Ei el Tower separate by taking advantage of the prede ned behaviour in web browsers. However, practically the 303 redirection of the W3C TAG and the hash convention leave the question of whether a resource is an information resource or noninformation resource indeterminate, since there is nothing to prevent 303 redirection from being used to redirect from one information resource to another information resource, and the hash convention is dependent on media types, being more often used for named parts in the document in HTML instead of as a shortcut for distinguishing non-information resources and their associated descriptions. 3.
suggested that there are fundamental disagreements about If an HTTP resource responds to a GET request with what precisely the di erence between an HTTP entity and a 2xx response, then the resource identi ed by that a \representation of a resource" are, and that this leads URI is an information resource; to widespread problems with caching implementations in
HTTP [?]. David Boorh has proposed an informal cateIf an HTTP resource responds to a GET request with gorisation of what can be identi ed by a URI, noticing the a 303 (See Other) response, then the resource iden- confusion between `naming' and `identifying' and even `deti ed by that URI could be any scribing' [?]. Hayes has long attempted to elucidate the funIf an HTTP resource responds to a GET request with damental di erence between the use of resources to access a 4xx response, then the nature of the resource is un- webpages and the use of a URI to refer to some non-Web known. accessible thing [?]. Furthermore, the use of URIs to refer to physical entities and the subsequent clari cation of the One concrete example would be an agent is trying to access direct reference position has led to the OKKAM project, a a URI that refers to the Ei el Tower itself, project to build a catalogue of `entity' URIs that is supposed http://dbpedia.org/resource/Eiffel_Tower. Upon attempt- to directly refer to physical entities [?]. This general line of ing to access that resource with a HTTP GET request on thinking has led to a number of workshops at conferences such as the World Wide Web Conference and the European Semantic Web Conference devoted to this topic [?, ?].
Within the W3C, there is an informal activity of the W3C TAG called the `Architecture of the Semantic Web' (AWWSW) that has for over a year attempted to decipher Web architecture, in part prompting by the need to model HTTP in RDF directly in order for HTTP transactions to be validated via EARL, the RDF-based Evaluation and Report Language used by the W3C to validate new W3C standards and describe test-cases [?, ?]. Yet, HTTP in RDF currently does not model the notion of `resource' except with a misuse of rdf:Alt, so it must be corrected by integrating an ontology of resources like IRW. While both EARL and the AWWSW are attempting a much more detailed and low-level description of HTTP transactions than we attempt, the lightweight IRW ontology described in this paper should allow speci cations like HTTP in RDF to directly address the notion of a `resource.'
Linked Data is to provide a foundation for the use of a common ontology to describe Linked Data and typical Linked Data transactions, currently being done by di erent ontologies in Section ??. To this aim, IRW can be discussed, reviewed, and comment on the ontologydesignpatterns.org wiki1. To serve the aim of elucidating arguments, additional modules of IRW have been developed and are brie y introduces in Section ??.
There have been previous attempts to model at least a subset of the notions outlined in a formal ontology, but all lack coverage of some crucial concepts. For example, while the ontology given by RDF Schema touches upon the vocabulary of resources via its term rdfs:Resource, it does not cover the distinction between information and non-information resources. The IRE (Identi ers, Resources, and Entities), based on Dolce Ultra Lite (DUL),2 a light version of the 1http://ontologydesignpatterns.org/wiki/Submissions:IRW 2http://www.loa-cnr.it/ontologies/DUL.owl widely-known DOLCE foundational ontology and its extension for describing information objects3 (IOL, described in [?]), attempted to model some of these concepts earlier [?]. However, many aspects were not included in IRE, such as the distinctions between resources and their Web representations, or the concept of accessing a web-page via a web server, that are crucial to the e orts within the W3C and Web community, while many of the distinctions drawn by DUL+IOL were found to be too `heavy-weight' for these communities [?]. In response to these concerns, the IRE ontology has been evolved into the IRW ontology. 5.
http://purl.org/NET/irw/ of the IRW ontology. The stable version of the ontology can also be accessed via its PURL. The latest version of the IRW ontology may be accessed at: http://ontologydesignpatterns.org/ont/web/irw.owl. The pre x rdfs: is used for the RDF(S) namespace http://www.w3.org/2000/01/rdf-schema#. ir: is http://www.ontologydesignpatterns.org /cp/owl/informationrealization.owl. While the IRW ontology in full can not explicated due to lack of space, the primary classes and properties are given in Figure ??. The IRW-related elements needed for the example of 303 redirection are given in Figure ??. The IRW ontology starts with irw:Resource. While this class expresses the same intuition as rdfs:Resource, we have de ned it because this version of IRW is within OWL-DL expressivity. In OWL Full, this class is equivalent to rdfs:Resource.
The notion of a URI is modeled as a class, irw:URI that has exactly one value for the datatype property irw:hasURI allowing to specify its value. Modelling URIs as a class allows us to talk about di erent kinds of URIs, such as IRIs (Internationalized Resource Identi ers) and Semantic Web 3http://www.loa-cnr.it/ontologies/IOLite.owl
Distinct from reference is the irw:accesses relationship, which is a causal connection to the thing identi ed. This is modelled again as a relationship between URIs and resources, although it is transitive, unlike irws:refersTo. If one can access a and a accesses b then a accesses c (via b).
Although a wide notion, access allows us to model the typical HTTP request-response Web transactions between a Web client and a server. A URI may also have a irw:redirectsTo property, a sub-property of irw:accesses, that we can use to model HTTP redirection. However, since redirection can URIs. According to some like Berners-Lee, URIs identify ex- be used between just information resources that have nothactly one resource. This is modeled in IRW by the functional ing to do with the Semantic Web, their domain and range property irw:identifies, having range irw:Resource (and says nothing about the type of resource. In order to model inverse property, irw:isIdentifiedBy). Of course, those explicitly the redirection, two distinct sub-properties of this that disagree with this viewpoint may not use irw:identifies, have been added in a TAG-speci c module of IRW4 that conand so it is given sub-properties irw:accesses and irw:refersTo. tains tag:redirects303To property and a tag:redirectsHashTo The idea of reference as explicated by Hayes is modeled property. Obviously, tag:redirects303To models the TAG's by the object property irw:refersTo (and inverse property, `solution' to httpRange-14 while tag:redirectsHashTo repirw:isReferencedBy) [?]. One condition on this property is resents the hash convention. that the object of reference should be \immediately causally disconnected" from its subject [?]. This is important, as reference is the relationship to both URIs for non-information resources like the Ei el Tower or integers, but also applies to the relationship of an information resource to some noninformation resource, like the relationship of Tim BernerLee's homepage to Berners-Lee himself. So, the key point is that URIs can identify resources, and some of these URIs refer to non-information resources.
Having de ned reference and redirection, we can now categorize resources. There are two main disjoint sub-classes of irw:Resource. The rst subclass is given as irw:InformationResource, which is an information object, such as a musical composition, a text, a word, a picture.
An information object is an object de ned at a level of abstraction, independently from how it is concretely realized.
So an irw:InformationResource expresses the same intuition and is an equivalent class to the DUL+IOL information object [?]. This means an information resource has, via the ir:realizes property (with inverse ir:isRealizedBy), at least one ir:InformationRealization, a concrete realization. This term is again imported from DUL+IOL [?].
So an information resource's \essential characteristics can be conveyed in a single message" implies that everything from a bound book to an HTTP message can be a realiza4http://www.ontologydesignpatterns.org/ont/web/tag2irw.owl associated with pre x tag:. tion for an information resource [?]. Furthermore, the prop- that relates a URI to a concrete Web server (inverse property irw:isAbout (and inverse property, irw:isTopicOf) ex- erty irw:resolvesTo). This irw:resolvesTo property is presses the relationship of an information resource to a re- currently implemented by mapping a URI to an IP address source or resources the information is `about.' Examples of or addresses. So each irw:WebServer is the resolution of this are descriptions of a resource using natural language or at least one irw:URI. Additionally, a irw:WebServer has a depictions of a resource using images. Information resources irw:isLocationOf property with at least one also can, but not necessarily, be identi ed (either accessed irw:WebRepresentation (inverse property, locatedOn), inor referred to) with a URI. In this manner, the text of Moby dicating the Web server concretely can respond to an HTTP Dick can be an information resource since it could be con- request with a particular Web Representation. veyed as a single message in English, and can be realized by both a particular book or a webpage containing that text.
Note irw:NonInformationResource complements Linked Data transactions.. irw:InformationResource from which it is disjoint with. The typical Linked Data transaction is also modeled. A Such class represents things that can not themselves { for new sub-class of irw:URI, SemanticWebURI is given, where whatever reason { be realized as a single digitally encoded the Semantic Web URI has a constraint that it must have message. A number of di erent kinds of things may be at least one irw:redirects property. In the Linked Data irw:NonInformationResources. Since this concept is the Initiative, another important kind of resource is \associated cause of much confusion and debate, it is detailed with three descriptions," which is just an Web resource that can be acdisjoint sub-classes. These kinds of IRW distinctions are cessed via redirection from a Semantic Web URI [?]. For not normative, as there are other possible plausible, more example, in DBPedia6 the resource detailed modeling choices. Our aim here is of communi- dbpedia:/resource/Eiffel_Tower redirects to an associcating the intuition behind the concepts of information and ated description at dbpedia:/data/Eiffel_Tower, and to non-information resources without entering the philosophi- an HTML page at dbpedia:/page/Eiffel_Tower dependcal debate about top-level ontologies. IRW contains three ing on the requested media type [?]. This scenario can be sub-classes of irw:NonInformationResources:5 generalized: irw:PhysicalEntityResource, is a resource that is `touch- a irw:WebClient irw:requests a irw:SemanticWebURI x able' like physical people, artifacts, places, bodies, chemical and the request is redirected (e.g. via hash or 303 redirecsubstances, biological entities; tion) to another URI, where this second URI identi es an irw:ConceptualResource, which refer to resources that are ldow:AssociatedDescription,7 which has one irw:isAbout created in a social process that can not be completely re- property to a non-information resource. We model alized digitally, such as legal entities, political entities, so- ldow:AssociatedDescription as a subclass of cial relations, as well as the concept of horse and imaginary irw:WebResource. objects like unicorns; and nally irw:AbstractResource, which refers to abstract combinatorial spaces that cannot be 6. ALIGNING IRW TO OTHER ONTOLOlocated in space-time such as formal entities like functions or the integers as well as more mundane resources like the in- GIES
nite set of names that constitute the resource identi ed by In this section, we present a number of suggested alignURIs themselves. A sub-class of irw:InformationResource ments, as given in Table 2. The alignments are to the three is irw:WebResource, which is an information resource iden- primary other ontologies, the RDF in HTTP ontology [?], ti ed by at least one URI and realized by at least one and the IRE ontology as well as an ontology for HTTP irw:WebRepresentation, so that a Web resource is just an used by the Tabulator Browser [?, ?]. The namespaces for information resource that is realized by at least one accessi- ont is http://www.w3.org/2007/ont/http. IRE, due to its ble Web representation like a web-page. irw:WebRepresentation modular construction and re-use of terms from DUL+IOL is a sub-class of irw:InformationRealization with con- patterns, uses many namespaces, but they can be found at straints added to make the cardinality of ir:isRealizedBy http://www.ontologydesignpatterns.org/cpont/ire.owl. and The http namespace is http://www.w3.org/2006/http#. irw:isIdentifiedBy both at least 1. In this way IRW can distinguish between a resource for the text of `Moby Dick' 7. APPLICATIONS in general and a webpage about `Moby Dick.'
The typical hypertext Web transaction can be modelled by IRW. We begin with irw:WebClient, which is some client in the context of the Web that can have a irw:requests relationship to a URI (note that irw:requests serves as an hook to the alignment of IRW with HTTP in RDF [?]), as exempli ed by a typical HTTP GET request). The irw:requests property is a sub-property of irw:access. A irw:WebClient then irw:requests a irw:URI. We also introduce the class irw:WebServer, which has a irw:isResolutionOf property 5Note that the three classes does not constitute an exhaustive partition.
is to solve the problem noted earlier that currently Linked Data resources are still not self-describing, such that there is no \de nition, description, some other kind of indication of what the identi er is intended to identify" on the level of a resource [?]. If one gets a URI of Linked Data, how can one record that it for a non-information resource or an associated description, besides actually going to the URI and performing HTTP GET. Then, how should one record 6Pre x dbpedia: is used for the namespace http://dpedia.org 7Typical Linked Data terminology is represented in a speci c module of IRW represented here by the pre x ldow: referring to the namespace http://ontologydesignpatterns.org/ont/web/ldow2irw.owl Class or Property irw:WebRepresentation
http:Content http:MessageHeader irw:InformationResource irw:SemanticWebURI irw:identi es irw:isAbout
Alignments owl:equivalentClass http:Message owl:equivalentClass ont:ResponseMessage rdfs:subClassOf ire:InformationRealization rdfs:subClassOf ir:InformationRealization rdfs:subClassOf ir:InformationRealization rdfs:subClassOf ir:InformationRealization owl:equivalentClass ir:InformationObject ire:SemanticWebURI ire:isExactProxyFor ire:about this provenance? The IRW ontology this in turn allows the semantic validation, to be able to describe and infer in detail the types of resources that can be interacted with via HTTP, which is useful for both tools like EARL that record validation of Web standards to be implemented in a reliable fashion, which is useful for error-reporting on the Web in general and HTTP in particular [?]. One facet of semantic validation is the description of Linked Data, where terms like non-information resource and associated description become important. This is useful for both semantic validation of Linked Data and Semantic Web Search engines [?]. 7.1
There would be a number of advantages if webpages that have RDF content could distinguish themselves as such, in the same way that HTML `valid' documents are currently validated by W3C Validators and often mark themselves by a computer graphic. This can be done by embedding a IRW statement in RDF/XML documents, RDF returned from SPARQL endpoints, and RDFa or GRDDL statement in XHTML or XML documents [?]. Ideally, this would be in conjunction with some sort of graphical logo to distinguish the page as `Linked Data Enabled,' as detecting the RDF statement, even in RDFa, is di cult for humans. Second, for irw:NonInformationResources that are part of the Linked Data and thus have no Web Representation to embed such a statement in, or resources whose actual Web can not be changed or must be changed en mass, such a RDF triple can be embedded directly in HTTP via the use of the HTTP Link Header [?]. 7.2
tect the presence of Semantic Web URIs and Information Resources, but also to determine constraints and contradictions. For example, one constraint that EARL is interested in nding out is whether namespaces documents are employing either the hash convention or 303 redirection, since according to the W3C, namespace resources are not information resources but an abstract space of in nite names. According to IRW, a namespace resource would be an irw:AbstractResource. This is because a user can `mint' a new namespace name without checking any namespace documents in any RDF and XML document and there is no ability of the namespace document to constrain names, but only to recommend them. One obvious use-case is to check every new namespace document and see if the namespace document can be reached through a irw:redirectsTo from a irw:NonInformationResource.
This same RDF records of what resources are Web resources or non-information resources, associated descriptions and their media-types (particularly RDF documents) is important information for any Semantic Web search engine. The proposed Semantic Web Site-maps allows authors to publish various characteristics of Semantic Web data, such as its update frequency and preferred method of access via an HTTP response [?]. However, it has to express what kind of data it is. This is important, as currently Semantic Web search engines often do specialise in di erent types of Semantic Web resources. For example, FALCON-S distinguishes between searching for what they call objects (irw:PhysicalEntityResources) and concepts (irw:ConceptualResources). As tools like Swoogle specialises in conceptual resources while the OKKAM project specialises in naming entities, by allowing publishers to describe what kinds of Semantic Web resources they have, a Semantic Web search engine can then specialise in searching and displaying for di erent kinds of resources [?, ?]. Furthermore, the use of a Semantic Web search engine that searches all kinds of RDF like Sindice, along with some large-scale inference engine like SOAR that could run some kind of inference-based reasoning algorithm against a large data-set, would allow the di erent kinds of resources to be automatically annotated and categorised [?, ?]. 7.3
validation. Currently, the only Linked Data validator is Vapour, which is coded procedurally and whose results can not themselves be presented as RDF [?]. The IRW and the HTTP in RDF vocabulary can be used to record whether or not each Linked Data resource is properly redirected using 303 redirection, and the IRW vocabulary can be used to make sure that the 303 redirection can lead access both an associated description in HTML and in RDF [?]. Any errors over large linked data-sets are easily collected and tested via SPARQL. Furthermore, Linked Data publishers could add two RDF statements that let their associated description be self-describing, solving the identity crisis in the context of Linked Data, and possibly leading to less incorrect use of owl:sameAs. Just embedding dbpedia:data/Eiffel_Tower irw:isAbout dbpedia:/resource/Eiffel/Eiffel_Tower would work. The following statement: dbpedia:data/Eiffel_Tower rdf:type ldow:AssociatedDescription could be added, as well as stating dbpedia:resource/Eiffel_Tower is of type irw:NonInformationResource or even irw:PhysicalEntityResource for clarity. This class would be useful for determining whether or not the resource had a property such as latitude or longitude, since concrete physical entities will have them while concepts and abstract mathematical expressions will not.
serve as foundational contribution of modelling Linked Data and so the \Dark Side of Semantic Web" that Hendler believes may give the Semantic Web a crucial advantage over previous e orts in knowledge representation [?]. IRW claries the interactions between the hypertext Web and Linked Data, allowing Linked Data spiders to keep track of important provenance regarding the identity of resources, and to characterise the resources correctly for semantic validation and error detection. Future work needs to be done to standardise IRW or a descendant thereof through the W3C, which will doubtless result in re nements to IRW, and to encourage its use within the Linked Data community in the context of various validators, debuggers, and search engines. By developing a consistent vocabulary for describing the identity of resources in IRW, the rst step has been taken.
We would like to thank Aldo Gangemi for his insightful comments. Also, Harry Halpin was partially supported by a Microsoft `Beyond Search' award. Valentina Presutti was supported by NeOn and IKS EU FP7 projects.