Cultural heritage on the Semantic Web: The Europeana Data Model

Resource Description Framework

The RDF is a model for describing resources and the relationships between them in a flexible and domain-independent way. Its basic structure is a three-element tuple (triple), subject → predicate → object, which allows sentences about resources to be made. An example is shown in Figure 1.

OPEN IN VIEWER

Figure 1.

Example of an RDF triple.

For the example in Figure 1, we can write the triple as follows: (<http://../queiros>,<http://../author>,<Os Maias>). The subject and predicate elements are Web resources, with a corresponding URI accessible via the HTTP protocol. The subject and object elements can have a literal value or a URI. The predicate element establishes a relationship or property between the subject and the object.

Multiple triples can be combined, building a network of unlimited information. For the triple shown in Figure 1, if we replace the literal value of the object with a URI, we get a labelled graph and, with multiple triples, we can represent more information (Figure 2) from a database, which is labelled ‘p’.

OPEN IN VIEWER

Figure 2.

Data representation (data set p).

A similar data set corresponding to a translation of the same work, database f, is shown in Figure 3. The existence of a common URI between the two structures makes it possible to link the two graphs and automatically obtain information about the original title of the work from database f, which is not in that database. One of the great advantages of Linked Data is that these databases can exist in different places on the Web.

OPEN IN VIEWER

Figure 3.

Evolution of RDF representation with additional information.

We can improve the merged result by adding some information considering the equivalence of the predicate naming: ‘p: author’ is the same as ‘f: auteur’. Moreover, both correspond to a resource category already defined in the community, such as ‘Person’ (‘foaf: Person’) ⁹ , with information about the name, date of birth and so on. The new representation of information, resulting from the addition of new phrases and existing terminology, allows the f-database user to ask questions such as ‘What is the web page of the author of the book?’, even if this information is not in their database. By connecting to an already defined data set (‘Person’), it is possible to combine this data set with other data sources, such as Wikipedia. ¹⁰ The resulting graph is shown in Figure 4.

OPEN IN VIEWER

Figure 4.

RDF representation with additional information and links to common vocabularies (Person).

To present RDF data, several serialization formats can be used, including RDF/XML, Turtle, N-Triples, Trig, N-Quads, JSON-LD and RDFa. Figure 5 shows an example of an RDF and its serialization in RDF/XML.

OPEN IN VIEWER

Figure 5.

Example of RDF/XML serialization.

RDF vocabularies, ontologies and SPARQL

To allow interoperability, the RDF representation of data must be meaningful and contextualized. The RDF Schema is an extension of the RDF that allows RDF vocabularies to be defined – that is, it provides the mechanisms for describing groups of related resources and the terms used to relate those resources (‘predicate’ in the RDF triple). In a simplified description, it is equivalent to a database schema, but in this case the definition is distributed across the Web, can be defined by anyone and will become common usage of Linked Data if it is used by many people on the Web (Wood et al., 2014). RDF vocabularies are sets of terms to describe resources. Each term can be a class or a property. Classes bring together resources with the same properties. Properties can be relationships (between resources) or attributes (characteristics). An example is shown in Figure 6.

OPEN IN VIEWER

Figure 6.

Example of vocabulary (class and property).

Vocabularies are defined in HTTP URIs and should follow the Linked Data principles, reuse existing vocabularies, and provide useful information on how the terms they define should be used. They usually have a prefix for the conceptual area they represent (namespace), which is commonly used to simplify RDF serializations. For example, the Dublin Core namespace is defined by the URI http://purl.org/dc/elements/1.1/ and uses the prefix ‘dc’. ¹¹ The RDF itself is defined by the URI http://www.w3.org/1999/02/22-rdf-syntax-ns# and uses the prefix ‘rdf:’. The lines in bold in Figure 5 show the URIs of the vocabularies used and their prefixes. In line 4 and line 6 of Figure 5, the formulations rdf: about and f: titre describe attributes of the resource defined by rdf: Description, whose identifier is given in rdf: about=“http://…/isbn/9782367321127” (line 4). In line 5, the attribute f: original defines the relationship between the described resource and the one with the URI identified through the attribute rdf: resource=“http://…/isbn/9781501146213”.

When different data sets use the same terms, vocabularies serve to resolve ambiguity or provide additional information that allows the discovery of new data relationships. Of great importance for the library community is the SKOS, an RDF vocabulary that aims to transform traditional knowledge organization systems (thesauruses, classification systems, subject headings or taxonomies) into Linked Data structures, and is concerned with maintaining compatibility with ISO 25964-2 (International Organization for Standardization, 2023).

Sometimes, the RDF Schema is not enough to guarantee all possible requirements, and the support of ontologies on the Semantic Web is necessary. An ontology is a knowledge organization model that defines a set of concepts and the relationships between those concepts for a given domain. It supports automated reasoning and data inference using logical rules, and allows knowledge to be shared and reused by people and machines. There are two W3C recommendations regarding ontologies: OWL and the Rule Interchange Format (Raafat, 2017).

To obtain data from the Semantic Web and use it to build applications or integrate it into web portals or web pages, it is necessary to have a language for searching and retrieving information: this is SPARQL, short for “SPARQL Protocol and RDF Query Language”. SPARQL can search local RDF (serialization) files or RDF databases anywhere on the Web, provided that these databases have a SPARQL endpoint service. ¹² The advantage of SPARQL and Linked Data is the possibility of combining local and remote queries to obtain the desired data (Wang and Yang, 2018).

Linked Data models in libraries

For several years, initiatives have been developed to foster the use of Linked Data in the library field. The Library of Congress, IFLA and Online Computer Library Center (OCLC), among others, have been committed to these efforts.

The IFLA developed the IFLA Library Reference Model (IFLA LRM), a standard that resulted from the consolidation of previously developed models for bibliographic information (Functional Requirements for Bibliographic Records (FRBR), Functional Requirements for Authority Data, Functional Requirements for Subject Authority Data), which were designed to support and promote the use of bibliographic data in Linked Data environments in the library universe (Joudrey and Taylor, 2018). Addressing print materials in more detail, this standard does not apply to many existing museum resources or presents difficulties for application in archival collections. This is one of the reasons that contributed to the development of specific conceptual models for these types of resources.

Based on the FRBR, Resource Description and Access was launched in 2011. It is structured for Linked Data applications and is a user-centred set of data elements, guidelines and instructions for creating metadata for library and cultural heritage resources. Resource Description and Access replaced the second edition of the Anglo-American Cataloging Rules (Park and Kipp, 2019).

Following this new model, the Library of Congress developed the Bibliographic Framework. This is an initiative to evolve bibliographic description standards towards a Linked Data model and to enable greater use of bibliographic information within and beyond the library community (Library of Congress, 2015). The Bibliographic Framework seeks to ensure the transition of all bibliographic heritage in MARC 21 ¹⁴ to the Semantic Web through a Linked Data model based on the RDF. In its current version (2.0), it is based on Linked Data technologies (the RDF Schema, the SKOS and OWL) and includes two main components: the Bibliographic Framework data model, which describes the structure of bibliographic resources and their potential relationship with other structures, and the Metadata Authority Description Schema in RDF, for knowledge organization. ¹⁵ In addition to developing the Bibliographic Framework, the Library of Congress (n.d.) makes available its ontologies, controlled vocabularies and other bibliographic description lists for use in Linked Data.

OCLC is another organization that is involved in research and development projects in the field of Linked Data. Its WorldCat database aggregates millions of bibliographic records from libraries around the world. It adopted the Schema.org vocabulary, which includes entities, entity relationships and actions, and was developed by Google, Microsoft, Yahoo and Yandex. ¹⁶ This allows WorldCat to expose its data for easier use in search systems and other applications. OCLC led the work to reconcile the Schema.org vocabulary with the Bibliographic Framework, and participated in the development of a Schema.org extension for bibliographic information. Also developed and maintained by OCLC, the Virtual International Authority File (VIAF) is an international authority file that includes authority records from national libraries around the world and has been available as Linked Data since 2009 (Wang and Yang, 2018). ¹⁷

Linked Data models in archives

For archival information, the Expert Group on Archival Description (2021) of the International Council on Archives is developing a conceptual model for archival description, which merges four standards for archival documentation (the General International Standard for Archival Description; the International Standard Archival Authority Records–Corporate Bodies, Persons and Families; the International Standard Description of Functions; and the International Standard Description of Institutions with Archival Holdings) and is called the Records in Contexts–Conceptual Model. Like the IFLA LRM for bibliographic description, this model identifies entities, the properties of entities and the relationships between entities. Associated with the conceptual model, a specific implementation was developed using OWL: the Records in Context Ontology. This enables archival description using Linked Data techniques. Both formulations are part of the Records in Contexts standard, which also includes Records in Contexts–Application Guidelines and Records in Contexts–Introduction to Archival Description. The most recent version of the Records in Contexts–Conceptual Model (0.2) is from July 2021.

Linked Data models in museums

The International Committee for Documentation (CIDOC) of the International Council of Museums has been working on a general data model – the CIDOC Conceptual Reference Model – to enable museum libraries, archives and other cultural heritage institutions to improve the sharing of information between each other (International Council of Museums, n.d.). It currently corresponds to ISO 21127:2014 – Information and documentation – A reference ontology for the interchange of cultural heritage information (International Organization for Standardization, 2020). According to Stein and Balandi (2019), there have been several attempts to provide a generally valid path for transferring data from Lightweight Information Describing Objects – CIDOC’s recommended XML schema for the delivery of museum metadata – to representations suitable for the Semantic Web. ¹⁸ The CIDOC Conceptual Reference Model is a complex system consisting of 94 entities (classes) and 168 relationships (properties) to represent and share cultural heritage metadata, and includes the representation of events associated with an object throughout its existence. Although its base structure is different from the models previously presented, its information can be mapped to other data models, such as the Dublin Core, Encoded Archival Description and the FRBR (Joudrey and Taylor, 2018).

Other models

In trying to align the CIDOC Conceptual Reference Model and IFLA FRBR and LRM, the new IFLA FRBRoo and LRMoo (object-oriented) models were developed (Bekiari et al., 2015). These include not only bibliographic entities, but also the activities related to their intellectual creation, editing and publication over time (Zapounidou et al., 2017). The Europeana Data Model, developed by Europeana, will be discussed in detail in the following section.

Europeana Data Model

Europeana aggregates digital objects, including works of art, books, photographs, letters, maps, newspapers, music and videos about art, archaeology, fashion, science, sports and more. They come from many repositories and use a wide variety of descriptive metadata standards. To provide access to digital cultural heritage and make it user-friendly, it was necessary to find a way to integrate these standards and make the metadata of all the partner institutions interoperable. For this purpose, a set of metadata that could encompass all this content with minimal loss of information had to be defined. To this end, the Europeana Semantic Elements Model was developed; this data model was used when the prototype of Europeana was launched in November 2008 (Ciocoiu, 2018).

The Europeana Semantic Elements Model is a linear structure that uses the Dublin Core Metadata Element Set with some Europeana-specific extensions. ²⁰ However, the weaknesses of the Europeana Semantic Elements Model were quickly identified – namely, the lack of distinction between the physical object in the institution’s collection and its digital representation; poor expandability; little flexibility to deal with the specificities of some objects; and the difficulty of creating semantic links between objects and resources. It became essential for it to evolve into a more flexible data model – the Europeana Data Model (EDM) – which is capable of overcoming the Europeana Semantic Elements Model’s problems and based on a Linked Open Data model. The Europeana Data Model is RDF-based and therefore an open and comprehensive model that allows links to other digital resources and data enrichment. Its development began in 2009 and is the result of the close collaborative work of Europeana experts in different fields with its internal team. The goal was to consider as many requirements as possible in the domains being worked on in an attempt to define a flexible model to represent data from the different sectors of Europeana. The data model was consolidated with some partners, such as the Deutsche Digitale Bibliothek and the Digital Public Library of America, which adopted the Europeana Data Model as the basis for their data models (Petras et al., 2017). ²¹ The more recent infrastructure ROSSIO, ²² ‘a free and open-access platform for aggregating, organizing, and connecting the digital resources in the Social Sciences, Arts and Humanities provided by Portuguese higher education and cultural institutions’ (Silva et al., 2022: 1), has adopted a Europeana Data Model profile that is capable of representing digital objects from different domains defined by Portuguese academic and cultural heritage institutions.

The Europeana Data Model solves the problems of the Europeana Semantic Elements Model in various ways: it distinguishes between the objects provided and their digital representations; it distinguishes object records from metadata records; it allows the representation of complex objects and for the detailing and expanding of the metadata format; ²³ it ensures compatibility with different levels of abstraction in the description; and it allows the use of contextual resources to incorporate concepts from controlled vocabularies.

Data representation

The Europeana Data Model defines its own vocabulary, containing specific elements (classes and properties), and reuses elements from other vocabularies such as the Dublin Core, Open Archives Initiative Object Reuse and Exchange, SKOS and Creative Commons. It also defines a set of descriptive and contextual properties that characterize the different aspects of a resource and relate it to other entities in its context. It can incorporate content with different description models (MARC 21, Encoded Archival Description, Lightweight Information Describing Objects) and with various levels of granularity. It also allows the same (physical) object to have several descriptions from different providers and to represent information added by Europeana.

The Europeana Data Model includes 11 specific classes and 6 classes belonging to other namespaces. For each class, it defines a set of properties, some of which are mandatory. In a simplified approach, a digital object provided to Europeana can be represented by three main classes and four contextual classes. Figure 7 is a graphical representation of a digital object from Europeana. The object is available at: http://www.europeana.eu/pt/item/08711/https___www_biodiversitylibrary_org_item_65402; the original document is available at: https://www.biodiversitylibrary.org/bibliography/24162.

OPEN IN VIEWER

Figure 7.

Digital object schema in the Europeana Data Model.

The class ‘Provided Cultural Heritage Objects’ (‘edm:ProvidedCHO’) refers to the original object, which can be a physical object (book, painting, photograph, map, drawing) or a digital original. All of the metadata describing the object must be part of this class. Each original object can have one or more accessible digital representations associated with it, some of which may be used as ‘thumbnails/previews’ and belong to the ‘Web Resources’ class (‘edm:WebResource’).

The class ‘ore: Aggregation’ links an original object to its digital representations described by a provider and included in Europeana, allowing the whole structure to be considered as a logical unit. Using the properties defined in the Europeana Data Model, each instance of the ‘ore: Aggregation’ class is related to a resource corresponding to the original object, ‘edm:ProvidedCHO’, through the ‘edm:aggregatedCHO’ property, and to one or more ‘edm:WebResource’ resources that are its digital representation(s). The instances of the ‘edm:WebResource’ class may have the following links (among others): ‘edm:isShownAt’, which refers to the full display of the object on the original website; ‘edm:isShownBy’, which refers to the main representation of the object that may be used to generate the thumbnail by the Europeana system; and ‘edm:object’, which also refers to the main representation of the object and normally has the same URL as ‘edm:isShownBy’. For subsequent representations, ‘edm:hasView’ should be used.

Also associated with the class ‘ore: Aggregation’ are three types of properties to identify the content providers: ‘edm:dataProvider’, which is mandatory, identifies the institution that is responsible for the original object and the information about that object; ‘edm:intermediateProvider’ is the intermediate organization that selects, collects or curates the data, and makes the information available in an aggregator; and ‘edm:provider’ is the provider from which Europeana obtains the information to load into its system.

Descriptive metadata can be attached to the original object, usually using the Dublin Core schema. Direct links are created between the described object and its characteristics, which can be expressed through simple text strings such as ‘dcterms: title’, ‘dcterms: subject’ or ‘dcterms: creator’ (see Figure 7). These links can also be made to other entities in other data sets. All of the entities have a unique identifier (URI) so that they can be referenced internally (Europeana Foundation, 2014).

Context classes

Some of the values of the descriptive metadata may be related to other entities, such as author, with more details associated with them. If this metadata is populated with URIs – that is, in accordance with the Linked Data principles – the system (Europeana) will create classes based on these URIs. To enable semantic enrichment, the Europeana Data Model provides context classes: ‘edm:Agent’ to represent people or organizations; ‘edm:Event’ for events; ‘edm:Place’ for geographical or location entities; ‘edm:TimeSpan’ for time periods or dates; and ‘skos: Concept’ for all the entities that comprise knowledge organization systems (thesauruses, classification systems, authority files, etc.). Figure 8 shows a new version representing the same model using context classes.

OPEN IN VIEWER

Figure 8.

Digital object schema in the Europeana Data Model with context classes for author and subject.

If we consider the information regarding the author of the book, in Figure 7, additional details about the author are not provided. This description can be improved by creating an explicit link (URI) between the book described and a resource that delivers such information – the Virtual International Authority File record on Darwin (https://viaf.org/viaf/735/) – by introducing the class ‘edm:Agent’ in the link ‘dcterms: creator’ (see Figure 8). This semantic enrichment of data brings huge benefits for the information search and discovery process. The contextual information added with the class ‘skos: Concept’ in the link ‘dcterms: subject’, like the value ‘Evolution’, leverages links to other entities and allows the use of all the information available in Wikipedia about each subject, including its translation into several languages.

Descriptive metadata: object-centric and event-centric approaches

The Europeana Data Model allows two approaches to descriptive metadata. In the object-centric approach, there is a direct link between the object and its characteristics, using literal values or links to other resources. This approach is the one used in Figures 7 and 8. The event-centric approach emphasizes the description of the different events in which the object was involved, providing the creation of a more complete network of entities that reflect the history of the object. The CIDOC Conceptual Reference Model, mentioned above, underlies this approach, which is adopted by some of the institutions participating in Europeana, particularly in the areas of museums and archaeology.

The ‘edm:Event’ class serves this purpose with three properties that relate to the resource being described: ‘edm:wasPresentAt’ – the event in which it was involved; ‘edm:happenedAt’ – the place where it happened; and ‘edm:occurredAt’ – the time when it occurred. To maintain consistency, Figure 9 shows a representation of the same digital object in an event-centric approach. Two events have been added – the publication of the first edition of the book (1879) and the new edition present in Europeana (1900-1979).

OPEN IN VIEWER

Figure 9.

Digital object schema in the Europeana Data Model using an event-based approach.

This approach can serve as a basis for digital storytelling.

Integration of resources from libraries, museums and archives

The Europeana Data Model allows the integration of data from multiple sources received in various formats and the mapping of metadata standards – for example, the MARC 21 format used in libraries; the Metadata Encoding and Transmission Standard for digital libraries; ²⁵ Encoded Archival Description, which is used in archives; and Lightweight Information Describing Objects, which is used in museums. To integrate this content into Europeana, these formats are mapped and converted to the Europeana Data Model using RDF/XML serialization. The data provided is normalized and semantically enriched through links to specific vocabularies.

To share digital collections with Europeana, the scope of the institution’s collections must be aligned with the Europeana Content Strategy (Scholz et al., 2017) and meet the technical criteria defined in the Europeana Publishing Guide. ²⁶ To publish digital content, it is necessary to find an aggregator, this being an organization that gathers data and makes it accessible through Europeana.

At the University of Coimbra, we have a vast collection of digital cultural heritage content available in our digital library, Almamater, which is shared with Europeana through our national aggregator, the Registo Nacional de Objetos Digitais, managed by the National Library of Portugal. ²⁷ Briefly, for each digital object, we register links to the corresponding item in the library catalogue, which will be converted into the ‘edm:ShownBy’ and ‘edm:ShownAt’ properties of the Europeana Data Model. The metadata from our library catalogue is converted from MARC 21 to UNIMARC and imported into the Registo Nacional de Objetos Digitais platform using the ISO 2709:2008 exchange format (International Organization for Standardization, 2008). The data is checked and sent to Europeana, processed and pre-published for validation. After validation, the data is finally published. Depending on the quality of the metadata and images, there are four tiers of criteria related to the quality of the content and three tiers to the quality of the metadata (for more information, see Scholz, 2015).

The University of Coimbra has approximately 1200 digital objects in Europeana, which link to the International Image Interoperability Framework viewer in Almamater to display the images and navigate the document structure. ²⁸ This is due to the fact that the International Image Interoperability Framework resource mapping is not yet available on the Registo Nacional de Objetos Digitais platform, but we hope to embed this presentation mode in our Europeana objects soon.

Europeana developments

The Europeana Data Model is a flexible and interoperable model for the description of cultural heritage resources. It is the result of systematic collaborative work and has the possibility to be expanded. This collaborative work has made it possible to accommodate the particularities of resources in specific areas through the development of extensions and adjustments that can respond to these characteristics. Europeana facilitates this process by supporting the community through the provision of detailed documentation on the Europeana Data Model, the developments being made, and the definition of the Europeana Data Model profiles (changes to the base model) to support future data integration services on the platform.

Some of the profiles that have been defined are the Sound Profile, to support the characteristics of sound objects and the metadata used to describe them; the International Image Interoperability Framework for the Europeana Data Model to provide guidelines for institutions for including resource descriptions in the International Image Interoperability Framework format in the Europeana Data Model metadata; and the Annotations Profile, based on the W3C Web Annotation Data Model recommendation (World Wide Web Consortium, 2017), which provides a solution for representing annotations so that they can be shared and used by different platforms related to Europeana. Storytelling is also an area in which Europeana is investing by developing specific methods and techniques together with its partners.

Europeana also offers a wide range of application programming interfaces (APIs), ²⁹ allowing the creation of applications using the rich cultural heritage available in Europeana: the Search API for simple searches; the SPARQL service to search Europeana’s structured metadata; the Record API to retrieve the metadata of a specific record; the Entity API to access information about subjects, people and places; the Annotations API to add information about existing items in Europeana; and, finally, there is also the Open Archives Initiative Protocol for Metadata Harvesting (OAI-PMH) service to retrieve larger volumes of information or aggregate Europeana’s repository (Europeana Foundation, 2008).