Paper 114 (In-Use track)

Shaping Device Descriptions to Achieve IoT Semantic Interoperability

Author(s): Aparna Saisree Thuluva, Darko Anicic, Sebastian Rudolph

Full text: submitted version

Abstract: The Internet of Things (IoT) promises easy integration of
connected physical devices at a large scale. For this purpose use of seman-
tic technologies are widely acknowledged, as they enable devices to un-
derstand the meaning of data instead of merely exchanging it. W3CWoT
Working Group is standardizing Thing Description (TD) as a machine-
readable interface of a thing. Further on, species seman-
tics for capabilities of things. Using these models it is possible to describe
devices. However, they do not constrain semantics of IoT devices. For
example, provides one capability for a class of devices.
Often same-class devices, produced by dierent manufacturers, dier in
certain feature or data they oer. In order to represent device variants we
propose to extend Capabilities with RDF Shapes. In our
approach, TDs for device variants are automatically generated. It also
enhances thing discovery, semantic interoperability, validation of TDs,
and accelerates IoT application development.

Keywords: Internet of Things; Web of Things;; RDF Shape Languages; Shape Expressions (ShEx); W3C Web of Things Thing Description; Semantic Discovery; Semantic Interoperability

Decision: reject

Review 1 (by Anna Fensel)

The paper proposes a solution for modeling capabilities and constraints of IoT devices. Having this information modelled may facilitate the interaction with the devices, and make them more actionable, e.g. enabling them to be activated in a manner similar as to activation of a semantic web service.
Given that there are really many IoT, appliance and device ontologies, it would be highly beneficial at the beginning of the paper to provide a very clear typical example scenario, demonstrating the advantages your work will bring, and what exactly that you model which is not available at other ontologies.
Now the coverage of the IoT, device and appliance semantic models is generally very high. Even complex scenarios including the details of an appliance as well as its context can be modelled solely with existing ontologies, e.g. see our example work addressing energy efficiency of fridges:
In the first half of the paper, there is on the other hand, quite a lot of repetition in the texts: some working groups and details of the overall context of the work are mentioned over and over again.
The exact scope of the work should be also defined clearer. 
The work is implemented, but an extensive evaluation is missing; in fact, the development would have a further evolution (i.e. this paper describes a development which is still in progress).
The presentation of the paper needs to be improved, the narrative should be more fluent. There are also a few typos:
-	Page 5: “model is alligned” should be “model is aligned”,
-	Page 10: “over specify capabilities” should be “overspecify capabilities”,
-	In the reference list, the reference number 7 is not complete. 
Generally, given the fact, that there are many relevant IoT models and scenarios being created and implemented now, the reference list should also be extended, to show that the state of the art overview has been comprehensive.
*** after rebuttal note ***
Thank you for replying the review comments.
I will remain with my opinion and the evaluation score.

Review 2 (by anonymous reviewer)

While the paper is well written and includes a sound motivation and approach it is not appropriate for the in-use track. It does not include details of use of the approach in practice or an evaluation. Furthermore it does not address the extent to which discoverability is improved which is a clear aim of the approach.

Review 3 (by Albert Meroño-Peñuela)

In this paper, authors propose a methodology and a prototype to generate Internet of Things (IoT) Thing Descriptions (TDs) that contain fine-grained, device specific feature specifications. TDs are important in IoT because they announce the features of a device across the network, its interactions, and its requirements over the data it consumes and produces. Therefore, TDs are at the core of enabling semantic interoperability when IoT devices exchange data on the Web. However, TDs convey these features via so-called Capabilities -- mainly modelled within,-- which are too generic for specific devices and adapt to device families only, failing to concretize specific features of concrete devices within these families. In order to address this, the authors propose to use RDF Shapes (ShEx) to model these device particularities, and to map them to JSON Schema. A prototype in JavaScript is introduced.
The paper addresses an important and relevant problem for the Semantic Web from a practical, real-life perspective. The techniques used (ShEx, JSON-LD, TDs) are novel and relevant for in-use scenarios, which means the paper is addressed at an appropriate community. The authors include a considerable number of links to external resources, code repositories, and example specifications. The modelling of  Capabilities, interaction patterns, and data seems reasonable according to the stated requirements.
However, the paper has several fundamental issues. Most importantly, the paper does not contain evidence whatsoever of the actual use of their proposed method and tool, or any assessment of how the proposed solution has been taken out of the lab, beyond a small prototype. Similarly, evidence illustrating how many different sectors the solution has been implemented in is missing, and only one example on air conditioners is shown. It is hence very hard to assess the actual impact of the proposed solution in real-life scenarios, mainly due to the lack of reporting on measurable impact features (e.g. number of users, online visitors, contributors to code/templates, etc.). The second missing key point is an evaluation, which could have been either qualitative (e.g. quotations from users reporting on satisfied/unsatisfied requirements), quantitative (e.g. what is the performance impact of generating the TD templates, compared to only using the existing generic Capability -- this could have provided insights of the costs in overhead that specifying fine-grained thing variants entails), or both. The quantitative analysis could have been especially interesting, as it could have provided some light regarding the expense/tradeoff at which the proposed solution comes. Another issue is that the paper fails at convincing the reader about their motivation of technological choices. For example, the only argument on why ShEx was chosen over OWL ontologies comes only at the end of the paper (“Before RDF Shape languages it was a challenge to constrain RDF graphs under the Open World Assumption of OWL”); while the reader keeps wondering about this choice since early on in the paper. Simiarly, the preference of ShEx over SHACL is not convincingly argued. It could have been interesting on hearing the authors’ take on these, as with e.g. in OWL a whole body of reusable ontology design patterns would have become immediately at their disposal. Finally, the presentation of the paper could be improved in several aspects: (a) the introduction could have used a number of citations to illustrate previous approaches and better motivate the proposed approach (e.g. [1]); (b) the problem definition (i.e. lack of mechanisms to customize/concretize TD’s Capability) is argued on and repeated way too much over the paper; (c) full shape dumps in the form of Listings break the layout and make the reading of the paper uncomfortable (for this, I’d suggest to include little excerpts only as examples, referring to the full dumps via links; and use Figures where possible); and (d) the references section does not seem to adjust to Springer’s template.
All in all, I think this paper deals with an important problem at the intersection of the Semantic Web and IoT, with a great potential in markets and real-life scenarios. Unfortunately, the paper contains no evidence of actual use, nor a proper evaluation. Besides an interesting combination of promising ideas, both method and tool seem to be early work that has to be yet carried out of the lab. I would like to encourage the authors to keep working on their method and prototype, and sincerely hope that beyond the points I raised previously they find useful advice to push this research forward.
** Reaction after rebuttal **
I want to thank the authors for their response letter. Although it is valuable that they contributed a SHACL implementation in the meantime to mitigate one of my concerns, the main issues of the paper (lack of evidence of use, lack of proper evaluation, and signs of early work) prevail, and hence I am keeping my score.

Review 4 (by Anna Tordai)

This is a metareview for the paper that summarizes the opinions of the individual reviewers.
The topic of the paper, a solution for modeling the constraints of IoT devices, makes it very relevant for the In-Use track and for this community. However, the reviewers note that the paper does not contain an evaluation, nor does it measure impact of the technology. In some cases the lack of detail in the description of the implementation raises questions. These comments lead us to conclude that this work is not mature enough in this state.
Laura Hollink & Anna Tordai

Share on

Leave a Reply

Your email address will not be published. Required fields are marked *