Provenance Description of Metadata Vocabularies for the Long-term Maintenance of Metadata

Provenance comes from French verb “provenir.” Provenance means source or history or derivation of an object, which can be work, data, etc. The provenance of a piece of data is the process that led to the piece of data in a computer system (Moreau, 2010). According to the W3C Provenance Working Group, provenance is a record that describes the people, institutions, entities, and activities involved in producing, influencing, or delivering a piece of data or a thing (Moreau et al., 2013). Provenance is used for many purposes, e.g. making judgments about information to determine whether to trust it, reproducing how something was generated (Gil et al., 2013).

Metadata vocabularies have to be maintained to keep metadata terms consistently interpretable. The definition of a metadata term may be changed, e.g. renaming of a term, revision of the meaning of the term, and revision of relationships to other related terms. It is crucial to trace changes of metadata terms in metadata vocabularies. Provenance description for long-term maintenance of metadata vocabularies is primarily the series of activities that have taken place on metadata vocabularies and their terms. This paper proposes a model to describe provenance description of metadata vocabularies based on W3C PROV. We classified entities and activities based on the relations defined in W3C PROV to describe primitive changes of metadata terms in metadata vocabularies. The recorded entities and activities are traceable to provide evidence for change tracking, which brings the benefits of provenance description of metadata vocabularies, e.g. preventing misinterpretation and auditing inconsistencies of metadata vocabularies. These benefits are valuable for the long-term maintenance of metadata vocabularies throughout their life cycle.

Provenance of metadata vocabularies is a record that describes the agents, activities, and entities involved in the lifecycle of metadata vocabularies. Provenance of metadata vocabularies includes information about how metadata terms in a metadata vocabulary and its term definitions come to a specific state. The definitions of metadata terms can change over time. For instance, a term can be split into two related terms, or the semantic relationship between two terms can change over time. Those who are responsible for maintaining metadata vocabularies need to pay attention to the changes and also document the changes.

Groth et al. (2012) illustrated requirements of provenance on the Web. The requirements refer to many dimensions, e.g. activities, records of changes, derivation, and interoperability. These requirements present the content of provenance and their use requirements. However, these requirements are not directly oriented to metadata maintenance. Keeping track of provenance of metadata vocabularies is beneficial to the consistent maintenance of metadata vocabularies. Provenance description of metadata vocabularies should be recorded in machine-readable, traceable and interoperable form to support the effective check of inconsistency caused by changes.

Machine-readability: to record provenance description of metadata vocabularies in machine-readable form for machine process, e.g. RDF/XML and RDF/JSON.

Traceability: to use provenance description of metadata vocabularies for tracking the changes among different versions of a metadata vocabulary, e.g. tracking provenance description in RDF using SPARQL

The SPARQL Protocol and RDF Query Language (SPARQL) is a query language and protocol for RDF. Please see the details at http://www.w3.org/TR/sparql11-query.

Interoperability: to keep provenance description of metadata vocabularies interoperable in the heterogeneous Web environments.

The reuse of existing metadata terms is essential to improve metadata interoperability. Metadata registry plays an important role in collecting and sharing metadata vocabularies to achieve metadata interoperability. Although metadata interoperability is an important aspect for long-term maintenance of metadata, metadata registry does not ensure metadata longevity. Metadata registry typically holds the following functions, i.e. registration, management, storage and sharing of metadata elements sets, and controlled vocabularies and application profiles. For example, Open Metadata Registry (OMR)

http://metadataregistry.org

OMR also provides service to vocabulary owners and managers about the versioning and change tracking of their registered vocabularies. The information about changed time, action, and the vocabulary maintainer who made the change are accessible on OMR history page. RDA vocabularies (element sets and value vocabularies) are maintained in the RDA Registry based on OMR with a combination of Git and GitHub. RDA Registry supports the semantic versioning of RDA vocabularies. The version designations follow the general principles of semantic versioning. GitHub provides the changes list of released RDA vocabularies in natural language, e.g. lists of “Adds new RDA entities,” “Adds new RDA elements,” “Adds new constrained RDA elements,” “Deprecates published RDA elements,” “Adds value vocabularies,” and “Renames value vocabularies” (Phipps, Dunsire, & Hillmann, 2015). However, these changes of RDA vocabularies are not kept interpretable to machines over time.

Javed, Abgaz, and Pahl (2014) proposed a layered change log model to record the changes of ontology using RDF triple-based representation. The changes are recorded using their own change metadata ontology and existing Provenance Vocabulary Core Ontology terms. Chawuthai et al. (2016) presented a logical model named Linked Taxonomic Knowledge (LTK) and LTK Ontology for preserving and representing changes in taxonomic knowledge for linked data. The changes in conception or in the relationship between taxa are preserved as events along with aspects of time, provenance, causes, and effects. A tool supporting version management of RDF vocabularies named SemVersion has been developed (Kendall et al., 2008). SemVersion provides structural and semantic versioning for RDF models and RDF-based ontology language like RDFS (Völkel & Groza, 2006).

Changeset vocabulary defines a set of terms (e.g. Addition, ChangeReason, and Removal) to describe changes between two versions of a resource description by using two sets of triples, i.e. additions and removals (Tunnicliffe & Davis, 2009). Changeset vocabulary represents changes to resource descriptions using RDF reification. An update is represented by a set of statements about statements and whether they are added or removed (Meinhardt, 2015). Changeset vocabulary is used by LCSH to describe the information of “Change Notes” of subject headings. The document-centric approved list of new headings and revisions to existing headings in LCSH are available on the Acquisitions and Bibliographic Access Web page

https://www.loc.gov/aba/cataloging/subject/weeklylists/

The W3C PROV standard for provenance description and provenance interchange is developed by W3C Provenance Working Group in 2013. The data model defined by W3C PROV, i.e. PROV-DM is used to encode the revision history of wiki pages (Missier & Chen, 2013). Getty Thesaurus of Geographic Names adopts W3C PROV to describe revision history of geographic names. W3C PROV is used to document the Activity information about the revision of geographic names, e.g. Activity type (Create, Modify) and temporal information associated with the Activity. Given to the extendibility of W3C PROV, this paper selects W3C PROV to record how metadata vocabularies change as provenance in RDF.

The W3C PROV standard includes a set of specifications which refers to many aspects of provenance, e.g. modeling, serialization, exchange, access, validation, semantics, and reasoning (Moreau et al., 2015). PROV-DM defines a conceptual data model along with relations to describe general provenance. PROV-O defines an OWL ontology consisting of a set of classes and properties for mapping PROV-DM to RDF. W3C PROV is for general provenance description and allows application to specific domains.

This paper applies W3C PROV to describe the provenance of metadata vocabularies. The main reason is that W3C PROV is a Web-oriented provenance standard for provenance description and provenance interchange. Entities and Activities are an important component to describe provenance in PROV-DM. An Entity is a physical, digital, conceptual, or other kind of thing (Gil et al., 2013). An “Activity” is something that occurs over a period of time and acts upon or with “Entities” (Moreau et al., 2013). An Activity can be used to represent how an Entity comes into existence, and how its attributes change to become a new Entity (Gil et al., 2013). To describe the provenance of metadata vocabularies based on W3C PROV, it is necessary to classify the Entities and Activities associated with changes among different versions of a metadata vocabulary. In other words, W3C PROV is used to describe the provenance of metadata vocabularies by defining what Entities have been changed and how the changes are caused by a series of Activities.

Vocabulary, Term, and Term Definition are classified as three subtypes of PROV Entity to describe provenance of metadata vocabularies. As illustrated above, a Term can be a concept or a class or a property. In the case of a concept, its definition may include its narrower term(s), broader term(s), association/related term(s), and other information. In the case of a class, its definition may include a description of its meaning, a label(s), a URI, super-class(es), sub-class(es), used property(ies), and other information. In the case of a property, its definition may include a description of its meaning, a label(s), a URI, super-property(ies), sub-property(ies), domain, range, expected value, and other information.

To describe the provenance of metadata vocabularies, Activities acting on the previously classified Entities are categorized into the following types, i.e. Revision, Addition, Deletion, and Replacement. Table 1 shows the correspondence of the classified Activities to the classified Entities. The mark “○” means “applicable” and “×” means “not-applicable.” Table 2 illustrates the classified Activities with their names and definitions. It is notable that replacement of term can be the following cases, e.g. a composite term was split into more than one term; or more than one term was merged to a term; or a term was replaced by another term. Table 3 provides change types of metadata vocabularies as well as their terms with specific examples, which are mainly from the changes between BIBFRAME 2.0 vocabulary (BIBFRAME 2.0 vocabulary list view, 2016) and BIBFRAME 1.0 vocabulary (BIBFRAME 2.0 specifications notes, 2016). In this paper, the separation of a single term into two or more terms is called a split. An example of a split in a subject heading is given in Table 3.

Table 1

Activities acted on Entities for provenance of metadata vocabularies.

Table 2

Definitions of the classified Activities for provenance of metadata vocabularies.

Table 3

Primitive change types of metadata vocabularies and their terms with examples.

A revision of a vocabulary is caused by a revision of its terms. The revision of a term may be a revision of the term as an instance, or a revision of documentation of the term. For example, replacement of a single term by a set of terms is a revision of an instance, and replacement of a title text is a revision of term definition. Therefore, the relationships between the classified Activities are as follows. A RevisionOnVocabulary is comprised of RevisionOnTerm (zero or more than one) and RevisionOnTermDefinition (zero or more than one). Given to the practical change examples of revision of a term and revision of term definitions, RevisionOnTerm has three general types, i.e. AdditionOnTerm, DeletionOnTerm, and ReplacementOnTerm; RevisionOnTermDefinition has three general types, i.e. AdditionOnTermDefinition, DeletionOnTermDefinition, and ReplacementOn TermDefinition.

The relations between Entities and Activities defined in W3C PROV include Usage, Generation, and Invalidation. Usage means utilization of an Entity by an Activity. Generation means creation of a new Entity by an Activity. Invalidation means destruction, cessation or expiry of an existing Entity by an Activity (Lebo et al., 2013). The properties prov:used, prov:wasGeneratedBy, and prov:wasInvalidatedBy defined in PROV-O are used to respectively describe Usage, Generation, and Invalidation. W3C PROV also defines Derivation between Entities. A Derivation is a transformation of an Entity into another, an update of an Entity resulting in a new one, or the construction of a new Entity based on a pre-existing Entity (Lebo et al., 2013). The property prov:wasDerivedFrom is used to directionally connect the two Entities from the new Entity to the pre-existing Entity.

Figure 1(a) provides provenance description in RDF graphs defined for the example of term replacement in Table 3: Subject heading “Folklore, Negro” is split into “Folklore, African” and “Folklore, Afro-American” (Knowlton, 2005). The classes and properties with prefix “mv” are defined in this research. The property mv:wasSplitTo is to describe the split of a term to more than one term. The class mv:Term is to assert a term of a metadata vocabulary as an instance of mv:Term using the property rdf:type. The class mv:ReplacementOnTerm is to assert an Activity as an instance of mv:ReplacementOnTerm using the property rdf:type.

Figure 1

This paper assumes the following URIs to describe the headings: “Folklore, Negro” with “http://id.loc.gov/authorities/childrensSubjects/sj96004706,” “Folklore, African” with “http://id.loc.gov/authorities/childrensSubjects/sj96004704,” and “Folklore, Afro-American” with “http://id.loc.gov/authorities/childrensSubjects/sj96004705.” An Activity instance of mv:ReplacementOnTerm made “Folklore, Negro” invalidated and generated two headings, i.e. “Folklore, African” and “Folklore, Afro-American.” In the split of a LCSH term, the Library of Congress Subject Headings Supplemental Vocabularies: Children’s Headings (LCSHAC) is a thesaurus that is used in conjunction with LCSH.

This paper identifies the thesaurus Entity before the split by URI “http://id.loc.gov/authorities/childrensSubjects/pv” and the thesaurus Entity after the split by URI “http://id.loc.gov/authorities/childrensSubjects/sv.” These thesaurus Entities are named LCSHAC PV and LCSHAC SV, respectively. Figure 1(b) shows the derivation from LCSHAC PV to LCSHAC SV. LCSHAC SV was generated by an Activity instance of mv:RevisionOnVocabulary and LCSHAC PV became invalidated by the same Activity instance. The class mv:Vocabulary is defined to assert a metadata vocabulary as an instance of mv:Vocabulary using the property rdf:type. The class mv:RevisionOnVocabulary is defined to assert an Activity as an instance of mv:RevisionOnVocabulary using the property rdf:type. The Activity instance of mv:RevisionOnVocabulary connects with the Activity instance of mv:ReplacementOnTerm through the property dcterms:hasPart, which is used to describe the inclusion relationships between Activities in this study.