=Paper=
{{Paper
|id=Vol-2489/paper4
|storemode=property
|title=Mapping Languages: Analysis of Comparative Characteristics
|pdfUrl=https://ceur-ws.org/Vol-2489/paper4.pdf
|volume=Vol-2489
|authors=Ben De Meester,Pieter Heyvaert,Ruben Verborgh,Anastasia Dimou
|dblpUrl=https://dblp.org/rec/conf/esws/MeesterHVD19
}}
==Mapping Languages: Analysis of Comparative Characteristics==
https://ceur-ws.org/Vol-2489/paper4.pdf
Mapping Language Analysis
of Comparative Characteristics
Ben De Meester[0000−0003−0248−0987]? , Pieter Heyvaert[0000−0002−1583−5719] ,
Ruben Verborgh[0000−0002−8596−222X] , and
Anastasia Dimou[0000−0003−2138−7972]
Ghent University – imec – IDLab,
Department of Electronics and Information Systems,
Technologiepark-Zwijnaarde 122, 9052 Ghent, Belgium
{firstname.lastname}@ugent.be
Abstract. RDF generation processes are becoming more interoperable,
reusable, and maintainable due to the increased usage of mapping lan-
guages: languages used to describe how to generate an RDF graph from
(semi-)structured data. This gives rise to new mapping languages, each
with different characteristics. However, it is not clear which mapping lan-
guage is fit for a given task. Thus, a comparative framework is needed.
In this paper, we investigate a set of mapping languages that inhibit
complementary characteristics, and present an initial set of comparative
characteristics based on requirements as put forward by the reference
works of those mapping languages. Initial investigation found 9 broad
characteristics, classified in 3 categories. To further formalize and com-
plete the set of characteristics, further investigation is needed, requiring
a joint effort of the community.
Keywords: Mapping Language · RDF graph generation
1 Introduction
RDF graph generation started as an ad-hoc process: hard-coded applications
generate specific RDF graphs, typically from specific data sources in specific
data formats. Shortly after, mapping languages were introduced, of which the
RDB to RDF Mapping Language (R2RML) is the first W3C recommended map-
ping language to generate RDF graphs from relational databases (RDBs) [1].
When a mapping language is used, the rules that specify the generation are de-
tached from the processor that executes them [8]. This improves interoperability,
reusability, and maintainability of the rules, the processor, and the entire RDF
graph generation process [13].
New mapping languages – or extensions to existing mapping languages –
are proposed to increase functionality or user-experience, and cater to different
?
Copyright ©2019 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
2 De Meester et al.
use cases. For example, mapping languages were proposed extending W3C rec-
ommendations, such as R2RML and SPARQL. The RDF Mapping Language
(RML) is proposed as an extension of R2RML to support heterogeneous data
sources [7, 8], and SPARQL-Generate is proposed as an alternative mapping lan-
guage, serialized using a modified SPARQL syntax [12].
This multitude of mapping languages allows to support more use cases. How-
ever, this multitude also defeats the mapping language’s purpose: interoperabil-
ity and reusability of current mapping rules and processors decrease. Currently,
mapping rules are incompatible across processors of different mapping languages.
The differences between mapping languages are currently not presented in a
unifying framework. Finding a suitable mapping language thus involves a labo-
rious process where end-users need to investigate the characteristics of different
mapping languages and interpret the differences. There is no clear interoperable
description of (non-)functional characteristics of different mapping languages.
In this position paper, we investigate a set of mapping languages of which
the reference works of those mapping languages claim complementary character-
istics. We align and present an initial set of comparative characteristics as put
forward by the reference works. These characteristics can then be used to com-
pare different mapping languages against each other. This work is not meant
to be complete, but to trigger discussion. Further and thorough investigation
is required. After presenting our reference works in Section 2, we discuss the
characteristics in Section 3 and conclude in Section 4.
2 Reference works
R2RML is the W3C recommended mapping language for describing RDF graph
generation from a relational database (RDB) [1]. It presents a language specifi-
cally designed as mapping language, serialized in RDF. Other recommendations
that allow to interpret a datasource as an RDF graph exist, such as JSON-LD [16]
and CSVW [17]. These recommendations target a single datasource type. How-
ever, the need for supporting different datasource types for different use cases
influenced the proposal of (i) other non-standardized mapping languages, (ii) ex-
tensions of existing mapping languages, and (iii) different notations.
We discuss RML(+FnO), an R2RML extension that supports heterogeneous
datasources and data transformations [5, 7, 8]; xR2RML, an [R2]RML extension
that supports collections and nested mappings [14, 15]; FunUL, an R2RML ex-
tension that supports data transformations [11]; SPARQL-Generate, an alterna-
tive mapping language based on SPARQL [12]; and YARRRML, an alternative
notation to, a.o., RML [10]. Specifically, we discuss the references of Table 1,
and the requirements they put forward.
This list is not exhaustive, however, we choose to discuss these works as their
reference work(s) list a set of requirements, and these sets of requirements are
(partially) complementary with those of the other mapping languages.
RML The RDF Mapping Language (RML) extends the R2RML recommenda-
tion to take into account heterogeneous data sources [7, 8]. Apart from allowing
Mapping Language Analysis of Comparative Characteristics 3
Table 1. The reference works discussed in this paper.
Mapping language Reference work(s)
(1) RML(+FnO) [5, 7, 8]
(2) xR2RML [14, 15]
(3) FunUL [11]
(4) SPARQL-Generate [12]
(5) YARRRML [10]
to specify how to generate the subject, predicate, object, and optionally graph
resources, RML allows for specifying the logical source that describes the iter-
ation over the data records (e.g., iterating over tabular data or JSON objects),
and specifying the data source that specifies the actual data connection (e.g.,
reading a local XML file or accessing a remote NoSQL endpoint). Later, RML
was combined with the Function Ontology (FnO) [2, 4] to include arbitrary data
transformations in the generation descriptions (RML(+FnO)) [5].
xR2RML An [R2]RML extension to include mapping functionalities targeted
at hierarchical (NoSQL) data structures [14, 15]. These advanced functionalities
include (i) nested term maps (for addressing hierarchical relationships), (ii) col-
lections and lists, and (iii) combining multiple query languages (for addressing,
e.g., a JSON record saved in an RDB).
FunUL An R2RML extension to include data transformations in the form of
JavaScript snippets, and support for CSV data sources [11]. The presented work
poses a set of eleven requirements for mapping languages, applied to generating
RDF graphs from CSV files.
SPARQL-Generate An alternative mapping language using a SPARQL-like syn-
tax to describe the mapping [12]. It has extensible support for different data
sources and data transformations. The presented work poses seven functional
and non-functional requirements.
YARRRML A notation using YAML1 to provide a user-friendly way to describe
mapping rules [10] which can be translated into RML (or other) mapping rules.
3 Characteristics
We discuss the following characteristics based on the aforementioned reference
works, and their posed requirements. We further classify as non-functional, data
source support, or functional characteristics (summarized in Table 2). For each
characteristic, we state which reference work posed the original requirement.
1
https://yaml.org/
4 De Meester et al.
Table 2. Summarizing the characteristics of discussed mapping languages, numbered
from 1–5. It is specified when the reference works of a mapping language claimed
(or refuted) a certain characteristic of another mapping language. For example, the
reference works of (2) – xR2RML – claiming that (1) – RML(+FnO) – has characteristic
DS1 is noted as 3(2). Statements claimed by the author of this paper are starred.
YARRRML’s characteristics except for NF1 and NF2 are taken from RML(+FnO), as
YARRRML mappings can be tranlated into RML(+FnO) mappings.
Language NF1 NF2 NF3 DS1 F1 F2 F3 F4
(1) RML(+FnO) 7(4) 3(4) 3(4) 3(2/3/4) RDF 3 3(4) 7(2) 7*
CSV (3/4)
XML (4)
JSON (4)
HTML (4)
RDF (4)
(2) xR2RML 7(4) 3* 7* 3 RDB 3 7* 3 3
NoSQL
(3) FunUL 7(4) 3* 3 ~* RDB 3 3 7* 7*
CSV
(4) SPARQL-Generate 3 3 3* 3 CSV 3 3 3* 3*
XML
JSON
HTML
Binary
$5 YARRRML 3 3* 3(1*) 3(1*) 3(1*) 3(1*) 3(1*) 7(1*) 7(1*)
Mapping Language Analysis of Comparative Characteristics 5
3.1 Non-functional characteristics
Non-functional characteristics are in relation to the user of the mapping lan-
guage. This user is either the end-user (creating the mapping), or the developer
(integrating the mapping language processor into his/her application).
NF1: Easy to use by Semantic Web experts (from SPARQL-Generate/YARRRML)
Whether the mapping language is “easy to use by Semantic Web experts” [12]
to describe the generation process or not. This characteristic is addressed by
SPARQL-Generate and YARRRML, which both provide a syntax that is either
“familiar” (SPARQ-like) [12] or “human-readable” (YAML) [10] to write rules.
The other mapping languages are described in RDF, without a specific notation.
NF2: Based on Semantic Web standards (from SPARQL-Generate) Whether the
mapping language “[integrates] with a typical semantic web engineering work-
flow” [12] or not, i.e., if it is related to existing standards. This characteristic is
fulfilled by all mapping languages, namely, they are related to R2RML, SPARQL,
or YAML.
NF3: Fully covering the generation process (from FunUL) Whether the map-
ping language “allows the serialization of the [generation] process for further
reuse” [11] or not: whether the generation description is fully covered by the
mapping language, or certain parts are hard-coded. xR2RML requires retrieval
of the data records as part of the hard-coded process. The other mapping lan-
guages allow to specify the connection to the physical data source.
3.2 Data source support characteristics
DS1: Supporting heterogeneous sources (from RML/xR2RML/SPARQL-Generate)
Whether the mapping language is focused on a single type of data source, or can
support multiple. FunUL targets tabular data sources. The other mapping lan-
guages have support for – and are extensible to – other data sources.
Although this is not tied to the mapping language itself, we further detail
which data sources are currently supported by the reference mapping language
processors.
3.3 Functional characteristics
F1: Supporting general mapping functionalities (from FunUL) Whether general
mapping functionalities are provided or not, namely, specifying “M:N relation-
ships”, “literal to IRI”, “vocabulary reuse”, “data types”, “named graphs”, and
“blank nodes” [11]. These functionalities include (i) generating subject, predi-
cate, object, and (optionally) graph resources, (ii) joining data, and (iii) spec-
ifying the used ontology and datatypes. This is typically supported by most
mapping languages.
6 De Meester et al.
F2: Extensible (from RML/FunUL/SPARQL-Generate) Whether additional func-
tionalities can be added or not, to “allow data to be manipulated and trans-
formed” [11]. RML supports this and proves it by combining data descriptions
and FnO, FunUL allows including JavaScript snippets, and SPARQL-Generate
relies on SPARQL 1.1’s extension functions.
F3: Supporting nested hierarchies (from xR2RML) Whether nested data records
can be joined or not, to “map data elements from rows as well as structured
values (nested collections [...])” [15]. This is different from being able to query
a hierarchical dataset: you can query a hierarchical dataset to return (and join)
only non-nested data records. Instead, this characteristic relates to whether the
mapping language itself can handle hierarchical data structures or not. Except
for xR2RML, all R2RML extensions use the tabular datamodel to join data.
xR2RML also supports combining query languages when a nested data record
has a different serialization than the parent data record. SPARQL-Generate is
assumed to support this, as it extends SPARQL, a graph-based query language.
F4: Supporting collections and lists (from xR2RML) Whether the mapping lan-
guage allows “to generate hierarchical values in the form of RDF collections or
containers” [15] or not. xR2RML has built-in support, SPARQL-Generate allows
to generate all triples required to generate compliant RDF lists.
3.4 Discussion
The difference in support of certain functional characteristics can help finding a
mapping language that is suitable for a certain use case. However, other charac-
teristics also influence the choice of a mapping language. For example, depending
on the use case, a different mapping language notation might be preferred. On
the one hand, a description in RDF can be less ambiguous, allowing more ac-
curate analysis of the generation description. On the other hand, a mapping
description serialized in RDF is arguably less user-friendly to edit and maintain
by users.
Further investigation is needed into the difference between serialization and
notation of mapping languages, and the underlying model and semantics they
employ. For example, comparing SPARQL-Generate with YARRRML: the latter
is designed to be human-friendly, and used to employ the semantics (and func-
tionalities) as proposed by RML(+FnO). The former is based on the SPARQL
language, but exhibits similar functionalities compared to RML(+FnO). RML,
xR2RML and FunUL extend the model of R2RML, which is based on a tabular
data model. SPARQL-Generate, following SPARQL, is based on a graph model.
Finally, it needs to be investigated on how to divide which characteristics
apply to the mapping language and which apply to the processors of that map-
ping language. For example, a mapping language can be extended to a specific
datasource, however, support of that datasource might not be easily achieved in
the processor. Another example is the automatic generation of metadata of the
generation process. This is easier when the mapping language is described in a
machine-understandable format, i.e., RDF [3, 6].
Mapping Language Analysis of Comparative Characteristics 7
4 Conclusion
In this position paper, we provide an initial investigation towards a comparative
framework for mapping languages. A more systematic review is further required.
It is apparent that this effort cannot continue without support from the larger
community. This work needs to be extended to consider a more complete range
of existing mapping languages. Choices made in this work – specifically, the clas-
sification of characteristics into non-functional, data source support, functional,
and the level of detail of different characteristics – need to be verified.
The resulting framework will allow for a clear division of which mapping
languages support which characteristics, and allows for end-users to more easily
find the right language for their use-case. More, development effort can be con-
solidated on missing features and filling the gaps, instead of spending time on
re-developing existing functionalities.
Further formalization and comparison between mapping languages can be
tested using a uniform set of test cases. Recent works are looking into test cases
sets for specific mapping languages [9]. This work can be extended to provide a
set of test cases across mapping languages. We expect this work to start a larger
discussion, and to provide a basis for a more complete and accurate comparative
framework. The recently started Knowledge Graph Construction Community
Group can be a crucial driver for further investigation.
Acknowledgements The described research activities were funded by Ghent Uni-
versity, imec, Flanders Innovation & Entrepreneurship (VLAIO), and the Euro-
pean Union. Ruben Verborgh is a postdoctoral fellow of the Research Founda-
tion – Flanders (FWO).
References
1. Das, S., Sundara, S., Cyganiak, R.: R2RML: RDB to RDF Mapping Language.
Working group recommendation, World Wide Web Consortium (W3C) (Sep 2012),
http://www.w3.org/TR/r2rml/
2. De Meester, B., Dimou, A.: The Function Ontology. Unofficial Draft, Ghent Uni-
versity – imec – IDLab (2016), https://w3id.org/function/spec
3. De Meester, B., Dimou, A., Verborgh, R., Mannens, E.: Detailed provenance cap-
ture of data processing. In: Garijo, D., van Hage, W.R., Kauppinen, T., Kuhn,
T., Zhao, J. (eds.) Proceedings of the First Workshop on Enabling Open Semantic
Science (SemSci). CEUR Workshop Proceedings, vol. 1931, pp. 31–38. CEUR (Oct
2017)
4. De Meester, B., Dimou, A., Verborgh, R., Mannens, E., Van de Walle, R.: An
Ontology to Semantically Declare and Describe Functions. In: Sack, H., Rizzo, G.,
Steinmetx, N., Mladenić, D., Auer, S., Lange, C. (eds.) The Semantic Web: ESWC
2016 Satellite Events, Heraklion, Crete, Greece, May 29 – June 2, 2016, Revised
Selected Papers. Lecture Notes in Computer Science, vol. 9989, pp. 46–49. Springer
(Oct 2016)
8 De Meester et al.
5. De Meester, B., Maroy, W., Dimou, A., Verborgh, R., Mannens, E.: Declara-
tive data transformations for Linked Data generation: the case of DBpedia. In:
Blomqvist, E., Maynard, D., Gangemi, A., Hoekstra, R., Hitzler, P., Hartig, O.
(eds.) Proceedings of the 14th ESWC. Lecture Notes in Computer Science, vol.
10250, pp. 33–48. Springer, Cham (May 2017)
6. Dimou, A., De Nies, T., Verborgh, R., Mannens, E., Van de Walle, R.: Automated
metadata generation for Linked Data generation and publishing workflows. In:
Auer, S., Berners-Lee, T., Bizer, C., Heath, T. (eds.) Proceedings of the Work-
shop on Linked Data on the Web co-located with 25th International World Wide
Web Conference (WWW2016). CEUR Workshop Proceedings, vol. 1593. CEUR,
Montreal, Canada (Apr 2016)
7. Dimou, A., Vander Sande, M.: RDF Mapping Language (RML). Unofficial draft,
Ghent University - iMinds - Multimedia Lab (Sep 2014), http://rml.io/spec.html
8. Dimou, A., Vander Sande, M., Colpaert, P., Verborgh, R., Mannens, E., Van de
Walle, R.: RML: A Generic Language for Integrated RDF Mappings of Hetero-
geneous Data. In: Proceedings of the 7th Workshop on Linked Data on the Web.
CEUR Workshop Proceedings, vol. 1184. CEUR (2014)
9. Heyvaert, P., Chaves-Fraga, D., Priyatna, F., Corcho, O., Mannens, E., Verborgh,
R., Dimou, A.: Conformance Test Cases for the RDF Mapping Language (RML).
In: Proceedings of the 1st Iberoamerican Knowledge Graphs and Semantic Web
Conference (KGSWC). Villa Clara, Cuba (Jun 2019)
10. Heyvaert, P., De Meester, B., Dimou, A., Verborgh, R.: Declarative Rules for
Linked Data Generation at your Fingertips! In: Proceedings of the 15th ESWC:
Posters and Demos (2018)
11. Junior, A.C., Debruyne, C., Brennan, R., O’Sullivan, D.: An evaluation of uplift
mapping languages. International Journal of Web Information Systems 13(4), 405–
424 (2017)
12. Lefrançois, M., Zimmermann, A., Bakerally, N.: A SPARQL extension for gener-
ating RDF from heterogeneous formats. In: The Semantic Web 14th International
Conference, ESWC 2017, Portorož, Slovenia, May 28 – June 1, 2017, Proceedings.
pp. 35–50. Springer International Publishing, Portoroz, Slovenia (May 2017)
13. Maroy, W., Dimou, A., Kontokostas, D., De Meester, B., Verborgh, R., Lehmann,
J., Mannens, E., Hellmann, S.: Sustainable linked data generation: The case of
DBpedia. In: d’Amato, C., Fernandez, M., Tamma, V., Lecue, F., Cudré-Mauroux,
P., Sequeda, J., Lange, C., Heflin, J. (eds.) The Semantic Web – ISWC 2017: 16th
International Semantic Web Conference, Vienna, Austria, October 21–25, 2017,
Proceedings, Part II. Lecture Notes in Computer Science, vol. 10588, pp. 297–313.
Springer, Cham, Vienna, Austria (Oct 2017)
14. Michel, F., Djimenou, L., Faron-Zucker, C., Montagnat, J.: Translation of Het-
erogeneous Databases into RDF, and Application to the Construction of a SKOS
Taxonomical Reference. In: International Conference on Web Information Systems
and Technologies. pp. 275–296. Springer (2015)
15. Michel, F., Djimenou, L., Faron-Zucker, C., Montagnat, J.: xR2RML: Relational
and Non-Relational Databases toRDF Mapping Language. Rapport de recherche,
Laboratoire d’Informatique, Signaux et Systèmes de Sophia-Antipolis (I3S) (Oct
2017), https://hal.archives-ouvertes.fr/hal-01066663/document/
16. Sporny, M., Kellogg, G., Lanthaler, M.: JSON-LD 1.0 – A JSON-based Serialization
for Linked Data. Recommendation, World Wide Web Consortium (W3C) (Jan
2014), http://www.w3.org/TR/json-ld/
Mapping Language Analysis of Comparative Characteristics 9
17. Tennison, J., Kellogg, G., Herman, I.: Generating RDF from Tabular Data on
the Web. Recommendation, World Wide Web Consortium (W3C) (Dec 2015),
https://www.w3.org/TR/csv2rdf/