W3C Workshop on Web Standardization for Graph Data

Position Statement  from

Gilles Privat, Orange Labs, France (gilles.privat@orange.com)

Abdullah Abbas, Orange labs, France (abdullah.abbas@orange.com)


on the topic :

Semantic graphs should not obliterate the preexisting use of graphs as models for physical systems.
They need to complement these graphs but should not subsume them.



“Cyber-Physical Graphs” vs. RDF graphs


Graphs have a long-lived and multifaceted history as models for all kinds of systems, processes and phenomena. Network science [1], emerging from the mid-1990s onwards as a far-reaching transdisciplinary field, has propelled an ultimate expansion of graphs as universal models in both natural and social sciences. Limiting ourselves to engineering, we focus here on a wide-ranging subset of these graph models, used to capture key aspects of (man-made) physical systems. These systems may be designed in a traditional top-down fashion (such as e.g., buildings, or industrial robots), or they may be “systems of systems”[2] that come closer to natural complex systems in that they get composed, at the relevant upper level of system description, in a dominantly bottom-up fashion (such as e.g., cities, or robot swarms). Graphs are highly relevant models for these systems because they may capture both their structure as relationships between their constituent subsystems, and their behaviors as relationships between their states. We propose to use the phrase “Cyber-Physical Graphs” when these graphs get further enriched by capturing the monitoring & control links that turn the underlying plant into a proper “Cyber-Physical System” (CPS)[1].

The extensive body of anterior knowledge on CPS graph models, as used for e.g. discrete event systems [3], may get occulted if graph models are seen from the exclusive vantage point of knowledge graphs “à la RDF”. Our position is that RDF graphs and CPS graphs should be associated and should not remain locked in two separate worlds, as they have been so far, mostly for lack of interpenetration between their respective communities. CPS graphs stand to benefit from being opened up in the linked data sense and “semanticized” in the semantic web sense, i.e. by assigning semantics to individual resources (typically nodes of these graphs, but also, as we shall see, their edges/arcs). This does not mean that CPS graphs should be reduced to RDF graphs, for 3 main reasons

1.                  CPS graphs stand on their own without RDF-style semantics. They do in fact have a different kind of semantics of their own, such as operational semantics applied to various kinds of state-transition graphs as behavioral models of dynamic systems, or a more obvious similarity-based semantics when a graph captures and matches  the structure of a physical network such as e.g. a power grid. These semantics apply to the graphs as a whole and are not reducible to the kind of “per-resource” semantics, which RDF is meant to describe

2.                  the RDF metamodel, which reduces, in a nicely parsimonious way, all graph arcs to predicates/properties, is too constrictive to properly capture graphs as models of  physical systems. : if an arc/edge of these graphs stands for a physical connection between two nodes, such as e.g. a power line in an electrical grid, or a pipe in a water distribution system, it should be entitled to have properties of its own to fully describe the underlying piece of physical plant , which an RDF graph cannot natively support without either  a cumbersome reification process, or making an edge into a properly instantiated resource, which means a node, which forsakes the modeling benefits of a graph edge matching a link in a physical network

3.                  CPS graphs need to be used by graph-theoretical algorithms that analyze key aspects of their overall structure : these can be either basic tools like evaluations of average path-lengths or degree distributions, or more sophisticated tools from spectral graph theory. RDF graph are totally unsuited to an analysis by such graph-theoretical tools because they obfuscate, flatten and dissolve the relevant graph structure by mixing structural arcs with mere property values and typing arcs that have no structural relevance whatsoever. This obfuscation is compounded by the need for reification or conversion of arcs into vertices as mentioned before, hiding the actual connectivity structure underneath an additional layer of transformation.


An adequate graph model should thus retain the full expressivity and native structure of classical CPS graphs in these three regards. Our position is that property graphs [3][5] are the best existing meta-model for capturing CPS graphs. Property graphs are a class of directed, labeled & attributed multigraphs, informally defined as the common denominator model of graph databases. They have so far lacked a strong theoretical grounding as they have emerged from the use of database practitioners as a compromise to retain familiar key-value or object primitives within a graph. Property graphs are by no means universal graph models : such models that would require resorting to hypergraphs (with n-ary edges), or even hypernodes (where nodes of a graph are themselves graphs), cannot be expressed with property graphs. Yet property graphs make it possible, crucially, to single out as relationships (and thus first class citizens of the meta-model) those arcs that represent actual physical linkages between physical entities, themselves represented as nodes. Mere properties (corresponding to OWL datatype properties) are directly attached to both entities and relationships as attributes would be in an object-based model. Paradoxical as it may seem, in a property graph, properties are actually not represented as arcs of the graph proper! This keeps the graph uncluttered and “clean” to represent saliently what matters the most: relationships as representations of physical connections between entities that make up the structural scaffolding of a system, and run graph-theoretical algorithms that analyze the structure of these systems.


Semanticization of a property graph should be seen as the superimposition of an overlay RDF/RDFS/OWL graph atop the property graph proper, whereby all entities, properties and relationships get assigned types derived from a formally defined ontology graph. The two graphs should be kept separate to avoid flattening and dissolving the structural property graph, which, as stated before, cannot be reduced to RDF-style semantics.


The meta-model defined by the ETSI CIM group [6] makes it possible to export property graphs to RDF in a standard way, by using reification through blank nodes to apply properties to relationships, and a proper set of RDF base classes to carry over the distinction between relationships and properties.

This gets the best of both worlds :

1.      the expressivity of property graphs, their adaptation to structural analysis and graph-based query tools, their native hosting on highly scalable and efficient graph data bases

2.      the universality and rich suite of reasoning and semantic query tools of the RDF model, its openness to the linked data cloud.




[1]            Brandes, U., Robins, G., McCranie, A., & Wasserman, S. (2013). What is network science?. Network science, 1(1), 1-15.

[2]            Maier, M. W. (1998). Architecting principles for systemsofsystems. Systems Engineering: The Journal of the International Council on Systems Engineering, 1(4), 267-284.

[3]            Cassandras, C. G., & Lafortune, S. (2009). Introduction to discrete event systems. Springer Science & Business Media.

[4]            M. A. Rodriguez and P. Neubauer, “Constructions from dots and lines,” Bulletin of the Association for Information Science and Technology, vol. 36, no. 6, pp. 35–41, 2010.

[5]            Margitus, M., Tauer, G., & Sudit, M. (2015, July). RDF versus attributed graphs: The war for the best graph representation. In Information Fusion (Fusion), 2015 18th International Conference on (pp. 200-206). IEEE.

[6]            CIM/NGSI-LD Data model Specification Preliminary Draft  https://docbox.etsi.org/ISG/CIM/70-Draft/006-MOD0/CIM-006-MOD0v002.docx


Background of Gilles Privat on the topic :


Dr G. Privat has been working in the Internet of Things, Cyber-Physical systems and their associations with semantic models since the early 2000s, at the time when “Ambient intelligence” was the catchphrase of choice when emphasizing the interplay between the physical and information worlds. He has been the leader of the Internet of Things chapter of  the FIWARE project and has been among the founding participants of the ETSI CIM ISG.


List of publications on the topic. :


Abdullah Abbas, Gilles Privat: Bridging Property Graphs and RDF for IoT Information Management. Scalable Semantic Web Knowledge Base Systems, co-located with 17th International Semantic Web Conference (ISWC 2018), Monterey, California, USA; 10/2018

Wenbin Li, Gilles Privat, José  Manuel Cantera, Martin Bauer, Franck Le Gall: Graph-based Semantic Evolution for Context Information Management Platforms. 2018 Global Internet of Things Summit (GIoTS), Bilbao, Spain; 06/2018, DOI:10.1109/GIOTS.2018.8534538

Wenbin Li, Gilles Privat, Franck Le Gall: Towards a Semantics Extractor for Interoperability of IoT Platforms. Global IoT Summit, Geneva; 06/2017, DOI:10.1109/GIOTS.2017.8016247

Wenbin Li, Gilles Privat: Cross-Fertilizing Data through Web of Things APIs with JSON-LD. European Semantic Web Conference, Workshop on "Services and Applications over Linked APIs and Data", Heraklion, Crete; 05/2016

Gilles Privat, Pascale Borscia, Marc Capdevielle, Laurent Lemke: Edge-of-Cloud Fast-Data Consolidation for the Internet of Things. International Conference on Innovation in Cloud Internet and Networks, Paris; 03/2016

Dana Popovici, Gilles Privat: Capturing the Structure of Internet of Things Systems with Graph Databases for Open Bidirectional Multiscale Data Mediation. The Second International Workshop on Large-scale Graph Storage and Management, Rome; 05/2015

Mengxuan Zhao, Gilles Privat, Eric Rutten, Hassane Alla: Discrete Control for Smart Environments through a Generic Finite-State-Models-Based Infrastructure. AmI 2014, Eindhoven; 11/2014, DOI:10.13140/2.1.4196.3202

Gilles Privat, Mengxuan Zhao, Laurent Lemke: Towards a Shared Software Infrastructure for Smart Homes, Smart Buildings and Smart Cities. EITEC, Berlin; 04/2014

Mengxuan Zhao, Gilles Privat, Eric Rutten, Hassane Alla: Discrete Control for the Internet of Things and Smart Environments. Feedback Computing; 06/2013

Gilles Privat: Extending the Internet of Things. Communications & Strategies, Digiworld Economic Journal n° 87, 3d Q 2012, pp101-119

Gilles Privat: Phenotropic and stigmergic webs: The new reach of networks. Universal Access in the Information Society 08/2012; 11(3):1-13., DOI:10.1007/s10209-011-0240-1

Gilles Privat, Norbert Streitz: Ambient Intelligence. The Universal Access Handbook, Edited by Constantine Stephanidis, 01/2009: chapter 60: pages 60.1 - 60.17; CRC Press Taylor and Francis Group., ISBN: 978-0-8058-6280-3

G. Privat, Zheng Hu, Stéphane Frenot, Bernard Tourancheau, "Iterative model-based identification of building components and appliances by means of sensor-actuator networks".,2nd Workshop on eeBuildings Data ModelsSophia Antipolis, 27 October 2011

Marc Lacoste, Gilles Privat, Fano Ramparany: Evaluating Confidence in Context for Context-Aware Security. Ambient Intelligence, European Conference, AmI 2007, Darmstadt, Germany, November 7-10, 2007, Proceedings; 11/2007, DOI:10.1007/978-3-540-76652-0_13

Thibaud Flury, Gilles Privat, Fano Ramparany: OWL-based location ontology for context-aware services. AIMS 2004, Artificial Intelligence in Mobile Systems; 09/2004


Background of A. Abbas on the topic:


Dr Abdullah Abbas is interested in research in semantic web ontology modeling (OWL/RDFS) and computational characteristics/complexity of queries related to RDF in particular and graphs in general. He has been since 2017 the rapporteur of the data model activity in the ETSI CIM ISG

List of publications on the topic.


Abbas A., Genevès P., Roisin C., Layaïda N. (2018) Selectivity Estimation for SPARQL Triple Patterns with Shape Expressions. In: Web Engineering. ICWE 2018. Lecture Notes in Computer Science, vol 10845. Springer,

Abbas A., Genevès P., Roisin C., Layaïda N. (2017) SPARQL Query Containment with ShEx Constraints. In: Advances in Databases and Information Systems. ADBIS 2017. Lecture Notes in Computer Science, vol 10509. Springer,


[1] We take CPS in a broad sense, not limited to industrial control systems , to include all kinds of systems  that get instrumented by the Internet of Things, even if this includes more monitoring than control, as is currently the case with most  IoT systems