The trend in the internet is to publish more data enriched with semantic descriptions (e.g., FAIR metadata) and contextual links (e.g., Knowledge graphs), evolving from connected facts to a complete knowledge about digital objects. Persistent identifiers (PID) play an important role in this environment by providing long-lasting and unique identifiers for these digital objects, enabling FAIRness, and facilitating disambiguation and connection between them. However, PIDs are static while digital objects can be dynamic, which constraints the scenario where PIDs can be adopted without potentially generating misinterpretation of the status of the digital objects. In this paper, we analyze the limitations of existing PID approaches and we discuss the needs for dealing with dynamic environments. This discussion drove our proposal of the Persistent and Time-unique Identifier (PTID), that adds temporal information into PIDs and has an hierarchical structure to store versions of dynamic digital objects. We also present the impact of using PTID to the maintenance process of knowledge graphs.
The unambiguous identification of digital objects (DO) is of utmost importance for data and
knowledge intensive tasks and the identifier is a fundamental component for a metadata to
FAIRly describe a DO. For this reason, academia and industry are joining forces to create rules
and infrastructures to better and persistently identify DOs, to publish them in registries, and
to promote the reuse of these PIDs to reference DOs. The role of a PID registry is to keep
ifndable and accessible information about a DO, including the URL address where this object
can be found. There are several schemes and technologies to build and store identifiers [
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). Science Graphs Interest Group, Data Description Registry Interoperability (DDRI), Scholarly Link Exchange (Scholix), etc.
With the advance of AI technologies and the adoption of FAIR principles, DO are becoming more complex and dynamic resources having several dynamic metadata describing them, impacting the description and identification of the whole DO. We focus our work on these complex and dynamic DO, and our analysis takes into account several perspectives of DOs, such as when a change occurred (i.e., time-stamps), how the change was detected (i.e., what data changed inside of the new version of the DO), how the change was characterized (e.g., the evolutionary relationships created to link the versions as well as the taxonomy used to label this link), and how to refer to the diferent status of the DO (i.e., to specific versions of the DO). Our case study is the maintenance process of knowledge graphs (KGs), thus, we are also interested on how to extend existing strategies for PID creation in order to facilitate our analysis and to propose better maintenance process for KGs.
Referencing a DO is not a new problem and several solution exist. The Digital Object Identifier
(DOI) 1 is the most widely used one. However, a DOI is not adapted for DOs that have versions
because: 1) it does not allow to link the versions; 2-) it does not indicate what is the current
version of a DO. In fact, DOI is advised for DO that are immutable over time. Extensions to
DOI were proposed, such as in Zenodo 2, which proposes an hierarchical structure composed
of “Version DOIs” and “Concept DOIs”. The former identifies specific versions while the latter
identifies a collection of DOIs. The Memento protocol [
For the problem of knowledge graph maintenance, we need a solution that combines the
advantages of each of these methods (the hierarchical referencing structure, temporal
information, the changes details, and the flexible content management). Building/updating knowledge
graphs of DOs requires to establish links between DOs’ metadata (including the metadata of
each version) and it is frequently done by linking DOs’ PIDs [
In the next section, we will introduce the persistent and time-unique ID (PTID) and we will discuss how PTIDs can be used to facilitate the maintenance of knowledge graphs.
Identifiers can have embedded information (also called “intelligent” identifiers) or not (“dumb”
identifiers). Good practices advise to avoid embedding information in the PID and only use
dumb identifiers. However, in some cases embedded information in the PID are necessary [
Our approach only use dumb identifiers and split the identifier in two parts: one for the collective entity and one for the version entity. This approach is named Persistent and Timeunique IDentifier (PTID) and it regroups the versioning, hierarchical, and distributed methods. The collective entity represents the abstract (and stable) part of the DO while individual versions are represented as “part_of” of the collective entity, and have the dynamic set of information. In our approach, the temporal information is added to the node (as an attribute) while the relationships do not have temporal information, but exist only during the period of validity of both (source and target) nodes. For instance, the analysis of the MeSH ontology (Medical Subject Headings) version 2009 and version 2010 shows that the concept number D001411 changed its label from Bacillus stearothermophilus into Geobacillus stearothermophilus. In the same period, new concepts were added to provide more precision on the definition of Bacillaceae. The changes are illustrated in figure 1 as an evolutionary relationship between concepts. Notice that, for readability reasons, we only add to the figure one specialization of the concept Bacillaceae.
If this scenario (figure 1) was represented by other cited approaches, we will notice that
the Zenodo approach will provide a PID to each version of MeSH, the memento protocol will
provide a PID to each versions of the concept D001411 (as they have a URL assigned, for instance,
http://id.nlm.nih.gov/mesh/D001411), the GIT system will track the changes in the MeSH files
(e.g., change label), and the ORCID ID approach will allow changing the label without changing
the PID. None of them will be able to completely represent this scenario. Looking deeper into
the problem, if we want to build a historical knowledge graph [
PID:D001406
Bacillaceae change label kind_of kind_of
Split into current inferred kind_of current
2010 Geobacillus stearothermophilus
PTID: 0000-0328
Geobacillus current PID:D056506
2010 PTID: 0000-0438
PID:D001411 kind of change was implemented between these versions. Other types of evolutionary links are also allowed. For instance, in 2010 the concept Bacillaceae was split into a more precise concepts (e.g., Geobacillus). Notice that the version entity of these concepts only have the creation date (by default this is a dynamic information) as properties, all other properties are static information and are part of the collective entity.
This approach avoids the duplication of stable information, but requires a clear definition of what is stable or dynamic information in the graph. The advantage of using KG to represent PTIDs is that each DO can use diferent criteria to define stable and dynamic information, and these criteria can change over time without afecting the other DOs. However, all versions of the same DO must respect the same criteria.
In brief, we showed that our PTID approach can be used when a change in a DO results into the creation of a new DO and when it does not. The PIDs are diferent for each DO’s version, removing ambiguities and giving to users the possibility to refer to the collective entity or to each version of the DO (ensuring findability and accessibility to historical data). The DO can be described by several properties (including time stamps) and these properties are duplicated only if they are dynamic ones (saving storage spaces). Having the time stamps also give the possibility to query for information based on a validity period or on a precise date. Finally, the type of changes can be added to the properties of the DOs’ versions as evolutionary links.
We also highlight that DOs can be created long time before having a PID. For instance, the
concept “Bacillaceae” was created in 1975, but it was added to MeSH (and got a concept number)
in 1999. Thus, KG builder can choose to add a snapshot of a DO from a specific date (diferent
from the creation date). To record this information, we added the possibility to have a timestamp
as a property of the node for the versions of the concept. The evolutionary relationships are
important information for the maintenance process of a KG. We can use it to infer or to predict
links between other concepts. For instance, if a collective entity was split into new specialized
entities, thus we can infer the “kind of” relationship between the versions of these entities
(as shown in figure 1). In other words, if we search for patients with Bacillaceae, we can also
add in the answer the patients that have Geobacillus. This example of analysis is part of our
ongoing work on KG maintenance [
PTID was designed to reduce the number of manual actions. To illustrate the impact of PTID on our approach, let’s assume that changes can only occur in the relationships or in the nodes of the KG. As a general rule, we adopt that a relationship has as origin and target collective entities (i.e., this relationship is valid for all versions of the DO). If this initial assumption is not true anymore (e.g., in 2010 the concept D001411 is not a “kind of” Bacillaceae anymore, it becomes a “kind of” Geobacillus). Thus, the relationship becomes a dynamic information and will link the versions of the concept D001411 to their target. The following triples replace the deleted relationship (D001411:PTID: 0000-0316, kind_of, D001406) and (D001411:PTID: 0000-0328, kind_of, D056506). The KG with this illustrative example is presented in figure 1, where the final version of the relationships are in green.
For changes in the nodes (e.g., in the label), we need to create a new version of the modified concept and copy all relationships from the previous version into the new one.
In this paper we analysed the limitations of using existing PIDs in dynamic environments such as the Web. This analysis focuses on environments where the information is represented through knowledge graphs and there is an intention of the KG manager to provide access to previous versions of the KG. Keeping snapshots of KGs stored in diferent datasets is not scalable, increase the time to find the information, and can be very expensive to store the data. Approaches that add hierarchical and/or temporal information to the PID are necessary to work with dynamic DOs, but to support the maintenance proceess of KGs we also need to have information about the changes. We proposed the persistent and time-unique identifier (PTID) approach that split the identifier of a DO into two components: one to identify the stable part of the DO and another to identify the dynamic part. We discuss how this approach contributes to overcome the limitations of existing PIDs and we illustrate how PTIDs supports the implementation of KG maintenance tasks.