1 Introduction
Maintaining the accessibility of collections for future generations is a central mission of libraries and other memory institutions. Metadata longevity should be ensured to keep the long-term accessibility of data collections. However, we are facing the difficulties in metadata longevity, such as the consistent maintenance of metadata, maintenance of metadata vocabularies and metadata terms, structural and syntactic features of metadata, metadata description rules, and so forth. This paper focuses on consistent maintenance of metadata vocabularies and metadata terms. This is because the changes of definitions of a metadata term may not always be recorded appropriately. The definition of a metadata term may include meaning and usage of the term, relationships to other terms, human-readable labels, and so forth. Metadata terms are usually defined as a set of terms, which is called a metadata vocabulary. This paper aims to propose a metadata model designed to keep track of the changes to definitions of metadata terms and metadata vocabularies.
In digital preservation standards, e.g. Open Archival Information System (OAIS)①http://www.iso.org/iso/catalogue_detail.htm?csnumber=57284 and PREMIS②http://www.loc.gov/standards/premis/, provenance of digital objects is a required component that has to be recorded for longevity of digital objects. As provenance of metadata is crucial for metadata longevity of such digital objects, how to formally and consistently describe the provenance of metadata over time is an important issue. Provenance of metadata schemas and provenance of metadata vocabularies, as well as provenance of metadata terms have to be consistently recorded over time. This paper focuses on provenance of metadata vocabularies and metadata terms. Provenance description of a metadata term is a record that describes the revision history of the metadata term. Provenance description of a metadata vocabulary is crucial as well. This paper applies W3C PROV③http://www.w3.org/TR/prov-overview/ to record provenance description of metadata vocabularies and their terms. The reason for adoption of W3C PROV is that it is developed as a standard for general provenance description and provenance interchange in a heterogeneous environment (Gil et al., 2013). W3C PROV has been commonly applied to specific domains, e.g. earth science and social sciences (Cuevas-Vicenttín et al., 2016; Lagoze, Willliams, & Vilhuber, 2013; Masó, Closa, & Gil, 2015; Missier & Chen, 2013; Tilmes et al., 2013).
The goal of this paper is to propose a model for formal provenance description of metadata vocabularies to keep track of primitive changes of their terms. The classified primitive change types can be applied to terms expressing either properties or classes of resources, i.e. both property vocabulary and value vocabulary.
The rest of this paper is organized as follows. Section 2 clarifies the meanings of Term and Term Definition in this paper. Section 3 presents requirements of provenance description of metadata vocabularies for metadata maintenance. Section 4 summarizes the related literature about metadata registries services and representation of changes. Section 5 applies W3C PROV to provenance description of metadata vocabularies. Section 6 provides a detailed description of the proposed model in this paper. The concluding remarks are given in Section 7.
2 Metadata Vocabulary and Terms
In the library community, commonly used metadata vocabularies are controlled vocabularies and metadata element sets (Hyland et al., 2013; Isaac et al., 2011), e.g. subject headings, authority files, Resource Description and Access (RDA)④（See http://www.rda-rsc.org and http://www.rda-jsc.org/archivedsite/rda.htmlfor details.） element sets, and RDA value vocabularies. A metadata vocabulary is a set of metadata terms. In this paper, we use “metadata vocabulary” as a generic concept that includes two types, i.e. property vocabulary and value vocabulary. A property vocabulary is a set of terms expressing attributes of a resource and relationships between resources, which is often called metadata element set, e.g. Dublin Core metadata element set⑤（http://dublincore.org/documents/dces/） and BIBFRAME vocabulary⑥（http://bibframe.org/vocab/）. A value vocabulary is a set of terms expressing classes of resources and encoding schemes of property values, e.g. Library of Congress Subject Headings (LCSH)⑦（See LCSH introduction at https://www.loc.gov/aba/publications/FreeLCSH/lcshintro.pdf.）.
To propose general provenance description model for tracking primitive changes of metadata terms in metadata vocabularies, this study defines “Term” and “Term Definition” as follows.
Term in a metadata vocabulary is an individual entity, which represents a concept, a property, a class, and a metadata vocabulary. For example, a subject heading in LCSH, property “dct:title,” class “dct:Agent,” and vocabulary encoding scheme LCSH are examples of terms. In this study, we use “Term” in both meanings of property vocabulary term and value vocabulary term.
Term Definition of a metadata term is a set of descriptions that defines features of the term. The features are the human-readable label(s) of the term, the meaning of the term, relationships between terms, usage of the term, and other information. Term Definition may be seen as a set of statements, each of which defines a feature of the term. For instance, “the broader term of Vehicles in LCSH is Transportation” is a Term Definition of Term “Vehicles;” “the label of term subject in Dublin Core metadata element set is Subject” is a Term Definition of Term “dc:subject.” The two examples of Term Definition can be respectively represented as RDF triples, lcsh:sh85142531 skos:broader lcsh:sh85137027 and dc:subject rdfs:label “Subject”@en. The lcsh:sh85142531 stands for “Vehicles” while the lcsh:sh85137027 stands for “Transportation.”
3 Provenance of Metadata Vocabularies
3.1 Definition of Provenance of Metadata Vocabularies
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.
3.2 Requirements of Provenance Description of Metadata Vocabularies for Metadata Maintenance
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 SPARQL⑧The 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.
4 Literature Review
This section discusses related works from the two aspects that are closely related to this study - metadata registry and representation of changes.
4.1 Metadata Registry Services for Metadata Interoperability
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)⑨（http://metadataregistry.org）, RDA Registry⑩(http://www.rdaregistry.info) and Dublin Core Metadata Initiative (DCMI)(11)(http://dublincore.org/) metadata registry(12)(http://dcmi.kc.tsukuba.ac.jp/dcregistry/) are typical examples of metadata registries, which provide search and browse services of their registered metadata vocabularies.
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.
4.2 Representation of Changes
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 page13（https://www.loc.gov/aba/cataloging/subject/weeklylists/）. The changes to the subject headings are provided together with the literal words like “ADD FIELD” or “DELETE FIELD.” Although Changeset vocabulary is applicable to describe changes of metadata vocabularies, the use of RDF reification will make the description of changes of metadata vocabularies complex.
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.
5 Application of W3C PROV to Metadata Vocabularies
5.1 Why Use W3C PROV
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.
5.2 Entities and Activities for Provenance Description of Metadata Vocabularies
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. |
| Subtypes of PROV Entity | Subtypes of PROV Activity | |||
|---|---|---|---|---|
| Revision | Addition | Deletion | Replacement | |
| Vocabulary | ○ | × | × | × |
| Term | ○ | ○ | ○ | ○ |
| Term Definition | ○ | ○ | ○ | ○ |
Table 2 Definitions of the classified Activities for provenance of metadata vocabularies. |
| Activity name | Definition |
|---|---|
| RevisionOnVocabulary | The revision of the contents or information of a metadata vocabulary |
| RevisionOnTerm | The revision of a term of the metadata vocabulary |
| AdditionOnTerm | The addition of a term |
| DeletionOnTerm | The deletion of a term |
| ReplacementOnTerm | The replacement of term(s) by other term(s) |
| RevisionOnTermDefinition | The revision of a term definition |
| AdditionOnTermDefinition | The addition of a term definition |
| DeletionOnTermDefinition | The deletion of a term definition |
| ReplacementOnTermDefinition | The replacement of a term definition by another term definition |
Table 3 Primitive change types of metadata vocabularies and their terms with examples. |
| Change type | Example |
|---|---|
| Revision of a Vocabulary | BIBFRAME 1.0 vocabulary is revised to BIBFRAME 2.0 vocabulary. |
| Revision of a Term | |
| Addition of a Term | Class bf:Note is newly defined in BIBFRAME 2.0 vocabulary. |
| Deletion of a Term | Property bf:otherEditionOf that was defined in BIBFRAME 1.0 vocabulary is deleted in BIBFRAME 2.0 vocabulary. |
| Replacement of a Term | Property bf:credits in BIBFRAME 2.0 vocabulary essentially replaces bf:creditsNote in BIBFRAME 1.0 vocabulary; Subject heading “Folklore, Negro” is split into “Folklore, African” and “Folklore, Afro-American.” |
| Revision of a Term Definition | |
| Addition of a Term Definition | The inverse property to property bf:absorbed is added in BIBFRAME 2.0 vocabulary. |
| Deletion of a Term Definition | The definitions of property bf:otherEditionOf that was defined in BIBFRAME 1.0 vocabulary is deleted in BIBFRAME 2.0 vocabulary. |
| Replacement of a Term Definition | The expected value of property bf:copyrightRegistration is corrected in BIBFRAME 2.0 vocabulary. |
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 ReplacementOnTermDefinition.
5.3 Relations Between the Classified Entities and Activities
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. Example of provenance description of metadata vocabularies in RDF. |
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.
6 Discussion
The goal of this paper is to define a model for provenance description of metadata vocabularies based on W3C PROV and RDF. To achieve this, we defined primitive change types of metadata vocabularies and their metadata terms as shown in Tables 1, 2, and 3. Following the proposed model, the provenance description of metadata vocabularies and their metadata terms can be recorded in RDF, which is machine-readable and traceable using SPARQL. Keeping change history of metadata vocabularies traceable by machines is important to keep numerous metadata consistently interpretable.
The proposed model can describe the revision history of metadata terms. As shown in Figure 1 (a), the subject heading “Folklore, Negro” (before the split) connects with “Folklore, African” and “Folklore, Afro-American” (after the split) through property mv:wasSplitTo. The proposed model can also describe the revision history of documentation of metadata terms. For instance, the meaning of term “soundContent” in the RDA element set was changed from “Relates to an expression to a presence or absence of sound in a resource other than one that consists primarily of recorded sound” to “Relates to an expression to a presence or absence of sound in a resource” (RDA sound content, 2016).
Not only provenance description of metadata vocabularies but also provenance description of structural features of metadata is crucial for the long-term maintenance of metadata. Related to this paper, our previous papers present models for provenance description of metadata schemas (Li & Sugimoto, 2014; Li, Nagamori, & Sugimoto, 2015). The practical use and service development of metadata provenance to facilitate long-term maintenance of metadata is left as the future research.
Figure 2. Example of provenance description of a metadata term in RDF. |
7 Conclusion
Provenance tracking is an important issue for the long-term maintenance of metadata vocabularies. Evidence of such provenance of metadata vocabularies enables consistent maintenance of metadata vocabularies. This paper proposes a model to formally describe provenance of metadata vocabularies, especially how metadata terms and term definitions (e.g. meaning and usage) change over time.
In this paper, the W3C PROV standard for general provenance description is applied to describe provenance of metadata vocabularies. We classified primitive change types of metadata terms in metadata vocabularies with specific examples. This study proposes a general model for provenance description of metadata vocabularies to track the primitive changes of metadata terms between different versions of a metadata vocabulary, e.g. split and merge of metadata terms and revision of meaning of metadata terms.
Author Contributions
This paper is part of research results of the Ph.D. Project conducted by C.Q. Li (lichunqiuaa@126.com, corresponding author) under the guidance of Professor S. Sugimoto (sugimoto@slis.tsukuba.ac.jp). Both authors participated in the development of the provenance model and writing of the paper. Both authors proofread and approved the final manuscript.