=Paper=
{{Paper
|id=Vol-1597/PROFILES2016_paper1
|storemode=property
|title=Qualitative Analysis of Vocabulary Evolution on the Linked Open Data Cloud
|pdfUrl=https://ceur-ws.org/Vol-1597/PROFILES2016_paper1.pdf
|volume=Vol-1597
|authors=Mohammad Abdel-Qader,Ansgar Scherp
|dblpUrl=https://dblp.org/rec/conf/esws/Abdel-QaderS16
}}
==Qualitative Analysis of Vocabulary Evolution on the Linked Open Data Cloud==
https://ceur-ws.org/Vol-1597/PROFILES2016_paper1.pdf
Qualitative Analysis of Vocabulary Evolution on
the Linked Open Data Cloud
Mohammad Abdel-Qader1 and Ansgar Scherp1,2
1
Christian-Albrechts University, Kiel, Germany
stu120798@mail.uni-kiel.de, asc@informatik.uni-kiel.de
2
ZBW – Leibniz Information Centre for Economics, Kiel, Germany
a.scherp@zbw.eu
Abstract. We analyse the evolution of vocabularies on the Linked Open
Data cloud. Based on the recent statistics of the LOD cloud, we have
selected the twelve most dominant vocabularies in terms of their use in
different pay-level domains. The number of versions we found for these
vocabularies range between 2 to 11. While some ontologies exist for more
than 10 years (e.g., FOAF) others are only online since a few years (like
DCAT). Our analysis shows that many changes occurred on annotation
properties. This reflects a need for more clarification of the terms, es-
pecially at early versions of the vocabularies. The majority of changes
in the vocabularies are due to changes in other, imported vocabularies.
Thus, there is a co-evolution of different vocabularies. This insight has
practical impacts to ontology engineers. They not only need to consider
the evolution of the vocabularies they directly use, but also those they
import and indirectly depend on.
1 Introduction
The Semantic Web main objective is to give data in the web a well-defined mean-
ing. Those meanings can be represented using ontologies, which can be defined
as a descriptive form for the concepts and items in a specific field or domain
and provides specifications for those items and its relations to other concepts
[1]. The ontologies are subject to change (evolve) over time for many reasons,
such as changes in the ontology’s domain, resolving errors appeared in the cur-
rent or previous versions of the ontology, changes in its external vocabularies
that are used to establish those ontologies, or any other reasons for updating the
ontology.
Creating a new version of an ontology requires processes to handle and man-
age multiple versions of that ontology [5]. Many research focused on analysing
the evolution of some ontologies, and does not focus on the core of establishing
those ontologies; the vocabularies. We mean by vocabulary, a collection of basic
terms (types and properties) that have a broad meaning. Those vocabularies
can be general (suitable for all domains) or specific (some domains or a single
domain).
Copyright held by the authors.
In this paper, we focus on analysing the changes that occurred on a selected
set of vocabularies. We analyse them from different perspectives and observe how
vocabularies are influenced by changes made in their dependency vocabularies.
Furthermore, we analyse the types of changes that occurred in vocabularies
terms (classes and properties). Those changes can be additions, deletions, modi-
fications, or renaming of vocabulary terms. Other changes such as splits/merges
can be considered as additions/ deletions processes.
We clarified the percentage of changes occurred on the examined vocabular-
ies that caused by the external vocabularies they depend on establishing their
vocabularies, and which vocabularies are depend on their terms on establishing
and evolving the vocabulary. All those analyses are useful for ontology engineers
when they are establishing a new ontology or updating an existing one by hav-
ing a clear idea about dependencies and relations between vocabularies to choose
vocabularies terms that meet their needs.
The remainder of the paper is organized as follows: In Section 2, we present
and discuss related work. We present our methodology for analysing vocabularies
in Section 3. In Section 4 we give an overview for the examined vocabularies. In
Section 5 we make a discussion of vocabularies evolution. Conclusion and future
work in presented in Section 6.
2 Related Work
Current research focused on measuring what are the changes in the LOD cloud,
but not how they are changed. For example, Dividino et al. [2] proposed a frame-
work to measure the evolution of the data in a dataset over time. They applied
their dynamics functions on 84 weekly snapshots from DyLDO dataset, which
results a number that can be used to represent how data in the dataset are
evolved. Furthermore, Dividino et al. [3] analysed the usage of vocabularies on
the LOD cloud over time and observed how they are changed according to their
usage. They analysed the combination of classes and properties that describe a
resource and applied their analysis on a dataset by taking 53 weekly snapshots
from DyLDO dataset. Over six months, Kfer et al. [4] observed the documents
retrieved from DyLDO dataset they created. They analysed those documents us-
ing different factors, their lifespan, the availability of those documents and their
change rate. Also, they analysed the RDF content that are frequently changed
(added or removed). Finally they observed how links between documents are
evolved overtime (either increased or decreased). To keep track what are the
changes happened when publishing a new version of ontology, Neubert [6] com-
pared the SKOS vocabulary versions files and found the differences between
them and then stored those differences into two separated named graphs; inser-
tions and deletions. Then he used the version history graph to link the insertion
and deletion graphs with vocabulary versions files. Walk et al. [8] studied and
analysed the user behaviour during editing ontologies to support the ontologies’
editing tools. They derived nine hypotheses to describe the users’ change be-
haviour, and then applied those hypotheses on four real-world ontology projects.
They found that the hierarchical structure hypothesis had the highest influence
on the editing behaviourr. Furthermore, Walk et al. [9] analysed the collection of
actions in the change-logs files that made by users in the collaborative ontology
engineering environment, to increase the quality of ontologies they design. They
applied Markov chains into the International Classification of Diseases (Revision
11) dataset. Zablith et al. [10] published a survey presenting an ontology evolu-
tion cycle, trying to gather researchers’ work in ontology evolution community.
Furthermore, they analyse the different approaches of each stage in the ontology
evolution process. They suggest to integrate the tools used for ontology evolu-
tion, and share the research in this field using Web portals, beside sharing some
common use cases that needs to evolve.
3 Methodology
Schmachtenberg et al.[7] published a report that provides a detailed statistics
about the LOD cloud. They analysed a subset of the Linked Data Web. The
subset is based on crawling seed URIs from the datahub.io1 dataset, BTC 2012
dataset2 , and public-lod@w3.org3 mailing list.
Based on their report, we selected the top used vocabularies in their crawled
subset of the LOD Cloud and have different available versions to download. Our
methodology can be expressed in the following steps:
– Selection criteria: We chose the vocabularies that have been used in more
than five datasets (0.49% of 1014 datasets used in the statistics [7]).
– Exclusion: We excluded the vocabularies that have been used in less than
five datasets because they are rarely used as shown in the previous statis-
tics. Furthermore, we exclude some of the upper level ontologies like RDF,
RDFS and OWL. For RDF and RDFS there are only one downloadable
version for each of them. For OWL vocabulary, we excluded it because all
our selected vocabularies and in all their versions use the same entities from
it, and they are five Annotation properties (backwardCompatibleWith, dep-
recated, incompatibleWith, priorVersion, and versionInfo) beside ”Thing”
class. Therefore, we decided not to include them in this paper.
– Selection result: Based on our criteria in selecting the vocabularies, we
examined 62 vocabularies that have been used in more than five datasets.
We found twelve vocabularies from the 62 vocabularies had more than one
version and can be downloaded. We tried to collect vocabularies that cover
all the topical domains in the LOD Cloud.
– Downloading: We downloaded the available versions for the selected vocab-
ularies using Linked Open Vocabularies (LOV) observatory4 and the official
1
http://datahub.io/dataset?tags=lod
2
http://km.aifb.kit.edu/projects/btc-2012/
3
http://lists.w3.org/Archives/Public/public-lod/
4
http://lov.okfn.org/dataset/lov
sites of some vocabularies. By using Protégé 4.3.01 , we extracted the differ-
ences between each version to capture the evolution of those vocabularies.
Table 1 shows the number of downloaded versions for each vocabulary, the
period from the first to latest version for those vocabularies, the evolution
duration in years/months, and the average number of changes per year.
– Analysing: We analysed the changes that occurred in different versions of
the vocabularies (creating, deleting, modifying, or renaming). Those changes
can be on classes, properties (with different types), datatypes, or individuals.
We analysed the changes using different classifications. First, we count the
number of changes for each type of entities, i.e., classes and properties. Sub-
sequently, we observed the percentage of internal changes versus external
changes. Internal changes are changes that occurred on the entities (classes
and properties) that originally introduced and developed by ontology en-
gineers of the examined vocabulary. On the contrary, external changes are
changes on the vocabularies’ entities from other vocabularies that are used
to establish the examined vocabulary.
Table 1: Number of downloaded versions for the examined vocabularies and their
evolution period, i.e., first appearance to latest version, sorted by the number of
datasets they were used in based on the State of the LOD Cloud report 2014.
In addition, the table shows to the evolution period in years/months and the
average number of changes per year
Duration in #Changes
Vocabulary #Versions Evolution period
years/months per year
FOAF 10 03.04.2005-14.01.2014 8 years & 10 months 30
DCterm 3 14.01.2008-14.06.2012 4 years & 5 months 26
SKOS 8 26.03.2004-18.08.2009 5 years & 4 months 44
Cube 4 27.11-2010-31.07.2014 3 years & 8 months 10
bibo 2 03.06.2008-04.11.2009 1 year & 5 months 13
DCAT 6 24.04.2012-31.05.2014 2 years & 1 month 47
GN 11 05.10.2006-29.10.2012 6 years & 1 month 34
SWC 3 27.02.2009- 11.05.2009 3 months 184
PROV 3 03.05.2012-11.01.2015 2 years & 8 months 106
AIISO 3 14.05.2008-25.09.2008 4 months 79
ORG 10 06.06.2010-12.04.2014 3 years & 10 months 70
Cal 2 07.04.2004-11.01.2015 10 years & 9 months 6
4 Overview of the Vocabularies from LOD Cloud
In the following figures and tables, we summarize the statistics of the selected
vocabularies and their different versions. According to the 1014 datasets crawled
1
http://protege.stanford.edu
in the State of the LOD Cloud report (Version 0.4, 08/30/2014) [7], the examined
vocabularies in our study were used in multiple topical domains.
Table 2 shows the topical domains for our selected vocabularies. In addition,
we show the percentage of the datasets crawled in [7] that use these vocabular-
ies. The table shows twelve vocabularies. Please remind that we selected those
vocabularies based on the availability of their versions (if they have) to down-
load. Any vocabulary that does not have versions, or their previous versions are
not available to download, are excluded from our analysis. We found that 65%
of the 62 examined vocabularies just have one version, and 15% of those 62 vo-
cabularies have more than one version but they are not available to download.
Therefore, we excluded them from our study. In Table 2, we can see that two vo-
cabularies (FOAF and DCterm) are used in more than 50% of datasets crawled
in [7]. In addition, half of the selected vocabularies are classified as cross-domain
vocabularies.
Table 2: Vocabularies according to their domains and dataset’s percentage based
on the State of the LOD Cloud report 2014 [7].
Vocabulary Domain Number of datasets
FOAF Cross-domain 701 (69.13%)
DCterm Cross-domain 568 (56.02%)
SKOS Publications/ Cross-domain/ Geographic 143 (14.10%)
Cube Government/ Geographic 114 (11.24%)
bibo Cross-domain/ Social web/ Media/ 62 (6.11%)
Publications/Life Sciences
DCAT User-generated content/ 59 (5.82%)
Government/ Cross-domain
GN Geographic/ Life Sciences/ Media/ Social web 27 (2.66%)
SWC Social web 27 (2.66%)
PROV Government/ Cross-domain 21 (2.07%)
AIISO Publications/ Life Sciences 17 (1.68%)
ORG Social web 14 (1.38%)
Cal Social web 9 (0.89%)
Fig 1 represent the total number of each change type, i.e. created, deleted,
modified, or renamed classes, object properties, data properties, annotation
properties, for all vocabularies we included in our analysis. From figure 1c, we
can observe that most of the changes are related to the modification changes
type, especially in object properties and classes. Also, the second most changes
are the creating changes type, mostly in the annotation properties (figure 1a).
Table 3 shows the external vocabularies that are used in establishing each
vocabulary in our study. From the external vocabularies listed in this table,
we can note two things: First, there are three vocabularies (FOAF, DCterm and
SWC) stuck on their external vocabularies list that used in establishing their first
edition until the recent one. Second, we can note that there are two vocabularies
(a) Total number of created entities
(b) Total number of deleted entities
(c) Total number of modified entities
(d) Total number of renamed entities
Fig. 1: Total number for each change type (Create, delete, modify, and rename)
for all versions of the examined vocabularies
(GN and ORG) have many changes on their external vocabularies list during
their evolution period. GN partially changed its external vocabularies five times
during six years, and ORG four times during approximately four years.
The percentage of internal changes compared to external changes in the vo-
cabularies versions are shown in Table 4. We calculate the total percentage for all
internal changes occurred through the vocabulary evolution overall its versions;
from first to latest version. We can see that three vocabularies with more than
90% of internal changes (DCterm, GN, and PROV). Furthermore, there is just
one vocabulary (Cube) with a percentage of internal changes less than 50%.
5 Discussion of Vocabularies Evolution
We observe that most of the vocabularies if they evolves, most of the changes
occurred on annotation properties for more explanations (metadata) to clarify
classes, properties or individuals. This can be expressed as a need for more
clearance for terms, especially between the early versions. Another observation
is that vocabularies are highly static w.r.t. the number of external vocabularies
used to establish and develop them. We can conclude that the topical domains
are fully covered with terms in the existing vocabularies, and if there is a need
of change, ontology engineers can modify the existing vocabularies.
In most of the examined vocabularies, the terms that changed were internal
terms, i.e. the terms that were created for the examined ontology by ontology
engineers, not the terms that are imported from external vocabularies during
establishing or evolving the vocabulary. On the other hand, some vocabularies
such as Cube, bibo, DCAT, and ORG are changed because some changes in their
external vocabularies occurred. For example, in the Cube vocabulary, the per-
centage of the internal change is 43%, and the remaining percentage of change
is caused by other external terms over its four published versions of this vocab-
ulary. The first version from the Cube vocabulary was published in 27.11.2010,
and some of its external vocabularies are DCterm and FOAF. DCterm pub-
lished its latest version in 14.06.2012, before the next version of Cube (was in
02.03.2013) was published.
Another example is the bibo ontology. Analysing ten versions from that vo-
cabulary, we conclude that 35% of changes are caused by external vocabularies.
bibo uses many external vocabularies such as DCterm and SKOS, and both of
them had versions between the two published versions of bibo (first version was
in 03.06.2008 and the latest version was in 04.11.2009).
Vocabularies such as DCterm, GN, and PROV are dependent on their own
terms, and most of the changes are made on those terms. For example, in the
GN ontology, we analysed eleven versions, and the percentage of internal changes
are 97%, which means that when the ontology engineers needs to change, they
change their own terms. Another observation in DCterm and PROV vocabularies
is that their external vocabularies as shown in Table 4 are small, and they are
organized as upper level ontologies.
Table 3: External vocabularies usage during examined vocabularies’ evolution
Vocab. External vocabularies Notes
FOAF dc/owl/rdf/rdfs/vs/wot/xml/xsd For all versions
DCterm dcam/owl/rdf/rdfs/skos/xml/xsd For all versions
SKOS dc/dcterm/owl/rdf/rdfs In first version
dc/dcterm/owl/rdf/rdfs/foaf/vs V. from 31.03.2005
to 29.08.2008
dcterm/owl/rdf/rdfs V. from 17.03.2009
to 18.08.2009
Cube dcterm/foaf/owl/rdf/rdfs/ For all versions
scovo/skos/void/xml/xsd
bibo Address/dc/dcterm/event/foaf/owl/rdf/rdfs/ V. 03.06.2008
skos/time/vann/vs/ wgs84 pos/xml/xsd
Event/foaf/ns/owl prism/rdf/rdfs/schema/ V. 04.11.2009
skos/dcterm/vann/xml/xsd
DCAT dcterm/owl/rdf/rdfs/xml/xsd In first version
dc/dcterm/dctype/foaf/owl/rdf/rdfs/ In other remaining versions
schema/skos/vcard/vann/voaf/xml/xsd
GN skos/owl/rdf/rdfs/xml/xsd V. from 05.10.2006
to 10.05.2010
cc/dcterm/foaf/owl/rdf/rdfs/skos/ V. 22.09.2010
wg84 pos/xml/xsd
dcterm/foaf/owl/rdf/rdfs/skos/xml/xsd V. 05.10.2010
cc/dcterm/foaf/owl/rdf/rdfs/skos/vann/ V. 14.02.2012
voaf/xml/xsd
adms/cc/dcterm/foaf/mrel/owl/rdf/rdfs/skos/ V. 29.10.2012
vann/xml/xsd
SWC bibtex/dc/dcterm/doap/foaf/geo/cal/misc/owl/rdf/ For all versions
rdfs/sioc/swrc ext/vcard/vs/wordnet/xml/xsd
PROV owl/rdf/rdfs/skos/xml/xsd In first version
owl/rdf/rdfs/xml/xsd In other remaining versions
AIISO cc/dc/dcterm/dctype/owl/rdf/rdfs/skos/vann/ In first two versions
xml/xsd
cc/dc/dcterm/dctype/foaf/owl/rdf/rdfs/skos/vann/ In latest version
vs/xml/xsd
ORG dc/foaf/gr/opmv/owl/time/rdf/rdfs/skos/vcard/ In first version
xml/xsd
dcterm/foaf/gr/opmv/owl/time/rdf/rdfs/skos/ V. from 08.10.2010
vcard/xml/xsd to 30.09.2012
dcterm/foaf/gr/opmv/owl/time/prov/rdf/rdfs/ V. 06.10.2012
skos/vcard/xml/xsd
dcterm/foaf/gr/owl/time/prov/rdf/rdfs/skos/ V. from 15.02.2012
vcard/xml/xsd to 12.04.2014
Cal dt/owl/rdf/rdfs/xml/xsd In first version
dc/dt/xhtml/owl/rdf/rdfs/xml/xsd In latest version
Table 4: Internal changes percentage
Vocabulary Internal changes percentage
FOAF 79%
DCterm 99%
SKOS 89%
Cube 43%
bibo 65%
DCAT 71%
GN 97%
SWC 86%
PROV 98%
AIISO 89%
ORG 73%
Cal 77%
The vocabularies in our study are different in their evolution period and the
number of versions published so far. In our study, we are trying to analyse the
evolution behaviour for those vocabularies. Some vocabularies such as FOAF,
SKOS, GN, and ORG have many versions; 10, 8, 11, and 10, respectively. GN
has eleven versions in the period from 05.10.2006 to 29.10.2012, and ORG have
ten versions in the period from 06.06.2010 to 12.04.2014. We think this large
number of versions is caused by their topical domain they are used in, especially
in social web ontologies.
The last observation is that the vocabularies used in publications, geographic,
social web, and government topical domains have the largest number of changes,
this is obvious from Fig 1 using SKOS, GN, SWC, FOAF, and PROV vocabu-
laries number of changes. For example, PROV added 29 classes and 62 object
properties, and modified 51 classes and 80 object properties during its evolution
period (Three versions published from 03.05.2012 to 11.01.2015). Another exam-
ple is SWC, the ontology engineers modified 110 classes and 25 object properties
through the three published versions.
6 Conclusion and Future Work
Analysing the change behaviour of vocabularies can help ontology engineers
in establishing new ontologies and evolve existing ones. This study can give a
clear view about ontology dependencies (External vocabularies), and the relation
between change in ontologies and their external ones.
Changes are mostly made for internal terms if they compared by external
terms percentage. Furthermore, most of the vocabularies have a static number
of external ontologies to depend on during the evolution period, and if they add
or remove some vocabularies, the number of those additions and deletions is
small.
Topical domains such as publications, geographic, social web, and government
have a high percentage for change if they compared with other domains.
As a future work, we will analyse the usage of vocabularies’ in the Dynamic
Linked Data Observatory (DyLDO) dataset. We will select the vocabularies
that have a version before and after a specific snapshot of a DyLDO crawl (the
first snapshot was in 06.05.2012, the latest is until today). Furthermore, we
will include the vocabularies not considered so far in this study. Establishing
a framework for ontologies’ concepts history tracking system will be useful for
ontology engineers.
Acknowledgement This work was supported by the EU’s Horizon 2020 pro-
gramme under grant agreement H2020-693092 MOVING.
References
1. Eder, J., & Koncilia, C. (2004, January). Modelling changes in ontologies. In On the
Move to Meaningful Internet Systems 2004: OTM 2004 Workshops (pp. 662-673).
Springer Berlin Heidelberg.
2. Dividino, R., Gottron, T., Scherp, A., & Grner, G. (2014). From Changes to Dynam-
ics: Dynamics Analysis of Linked Open Data Sources. In PROFILES14: Proceedings
of the Workshop on Dataset ProfiIling and Federated Search for Linked Data.
3. Dividino, R. Q., Scherp, A., Grner, G., & Grotton, T. (2013). Change-a-LOD: Does
the Schema on the Linked Data Cloud Change or Not?. In COLD.
4. Kfer, T., Abdelrahman, A., Umbrich, J., OByrne, P., & Hogan, A. (2013). Observing
linked data dynamics. In The Semantic Web: Semantics and Big Data (pp. 213-227).
Spring-er Berlin Heidelberg.
5. Klein, M. C., & Fensel, D. (2001, July). Ontology versioning on the Semantic Web.
In SWWS (pp. 75-91).
6. Neubert, J. (2015): Leveraging SKOS to trace the overhaul of the STW Thesaurus
for Eco-nomics, In: Proceedings of the International Conference on Dublin Coreand
Metadata Appli-cations 2015, So Paulo (Forthcoming)
7. Schmachtenberg, M., Bizer, C., & Paulheim, H. (2014). State of the LOD Cloud
2014. URL: http://linkeddatacatalog. dws. informatik. unimannheim. de/state.
8. Walk, S., Singer, P., Noboa, L. E., Tudorache, T., Musen, M. A., & Strohmaier,
M. (2015). Understanding how users edit ontologies: comparing hypotheses about
four real-world pro-jects. In The Semantic WebISWC 2015 (pp. 551-568). Springer
International Publishing.
9. Walk, S., Singer, P., Strohmaier, M., Helic, D., Noy, N. F., & Musen, M. A. (2015).
How to apply Markov chains for modeling sequential edit patterns in collaborative
ontology-engineering projects. International Journal of Human-Computer Studies,
84, 51-66.
10. Zablith, F., Antoniou, G., d’Aquin, M., Flouris, G., Kondylakis, H., Motta, E., ...
& Sabou, M. (2015). Ontology evolution: a process-centric survey. The Knowledge
Engineering Review, 30(01), 45-75.