|Review Comment: |
The paper describes WSML-PE, a language for modeling pervasive environments using semantics to better support interoperability, scalability and composition of services in a IoT context. The key idea is to extend the WSML language, introduced to model Semantic Web Services (SWS) and add those features required in the new context. The formal semantics for the language is based on Z-language. The feasibility of the approach is discussed using a case study that shows how services can be modeled and (automatically) composed using the proposed approach.
The main contributions of the paper to the state of the art can be summarized as follows (- with some remarks with respect to the paper claims):
• The paper describes an extension of the conceptual syntax of WSML; the extension does not change the logical language behind, but, in practice, extends the ontology of the concepts considered as first class citizens in the language. The main contributions of this extension include the distinction between services and service instances, service hierarchies, special ontologies for modeling context (also with real time specifications), specialization of goals, choreographies and orchestrations.
• The use case shows an interesting scenario, which may open a renewal of interest around SWSs. In this domain (maybe more than in other traditional WS contexts), the need to bridge the gap between users’ conceptualizations and underlying systems so as to let users program their pervasive environments is well motivated.
• The paper is quite convincing about the advantages of the proposed solution in terms of scalability and reusability of pieces of the model (e.g., descriptions of class of services and contexts, when a new services is added)
• WSML-PE uses WSML but the authors claim they could import OWL ontologies (more details about this import would be useful)
• The paper shows an example of service composition using WSML-PE. The example shows that existing algorithms can be used on top of the proposed language.
However, the paper describes a research at a too early stage of development, which is reflected in the lack of a convincing empirical evaluation: while the paper explains how the language can be used to model a pervasive environment, evidence that the proposed language is effective in comparison with other solutions is not provided. Overall, I find the use case useful to understand the proposed approach but not conclusive about its effectiveness to solve a real-world scenario.
I summarize my main observations here bellow.
The WSML extension is reasonably well motivated, but it does not address any particularly challenging problem. New classes of objects are added, but the language and approach used to model services and link these models to background knowledge are very similar to the ones used in WSML (rules, preconditions, effects, etc.). The use of real-time constraints and of attributes (e.g., AttributePE) are among the novelties of the proposed model, but their impact on the model is not discussed in depth in the paper.
**Evaluation and related work
The evaluation fails to be convincing in demonstrating the effectiveness of the proposed language. The evaluation shows that the proposed language and approach can be used to model and compose services, thus the proposed solution is conceptually feasible and motivated (i.e., ontologies can be reused so as to reduce the effort needed in modeling a new service). However:
• The evaluation does not address the core of the interoperability problem. To convince that the solution can really be applied to real-world problems, a deployed system in a pervasive environment that uses real sensors and services and run with the proposed solution should be described as a demonstration of effectiveness. Else, many questions remain unanswered. For example:
1) How to bridge the gap between local data models and the WSML language? What about de facto web standards such as APIs and hREST?
2) How message passaging is supported at runtime?
3) Which is the effect of real time consideration on a real-world use case?
4) While I understand that many ontologies can be reused (I appreciated when you mention that ontologies are extracted from existing ontologies), which is the effort needed to generate the semantic descriptions and to interoperate with web protocols becoming more and more adopted (e.g., hREST)?
• Although it may be difficult to answer to all these questions in one paper, it is possible to focus on some of these problems and provide empirical evidence that the proposed approach solve it. For example, the authors may link their representations to sensor data and show that the system is able to react in a timely fashion. Or they may compare the effort needed to build/modify the integration of a set of sensors/services with the proposed solution vs a custom approach (e.g., what if a new sensor is added to the room, etc.).
• In relation to Questions 1 and 4, it is important to remark that the uptake of SWS models such as OWL-S and WSML was very limited outside of the research community. One obstacle to greater uptake of SWS models was found in the burden of creating complicated descriptions of service behavior. And in fact, part of the scientific community that has worked with SWS and with WSML then attempted to propose lighter SWS models (e.g., WSMO-Light, Linked USDL, and SAWSDL) that could be compliant with hREST and APIs, and to develop approaches for the semantic annotation of the latter models to extract semantic representations without requiring too much effort from the users (see annexed bibliography). Although it may be the case that those approaches fail to deliver a comprehensive solution for pervasive environments, it is reasonable to question why 1) a comparison with this more recent body of work in SWS is missing and 2) the motivations that lead many researchers to look into these directions do not apply in your scenario. Overall the question would be: why solutions that have been dropped in the WS field should be instead used in this scenario?
Runtime vs Design Time. While the composition is based on backtracking (from the goal to a composition of services that achieves this goal), I would expect that, in a IoT context, execution is triggered by sensor data. So I would expect to see a modeling phase and an execution phase. It is not clear if you assume that (a) services are discovered, composed and orchestrated at runtime, or that (b) orchestration is built at design time and then the system runs the workflow that orchestrate the services at runtime. From Section 6, I think that you propose (a) but this issue requires better explanation. Overall, this may be an important topic to focus on when proposing to adapt models used in “traditional” web services to a IoT context: while services are invocated upon user requests, some services should be triggered by sensor data in IoT.
Goal descriptions. While, for SWSs, it is clear that goals describe the effect that a user want to obtain when a WS is considered, I found this approach to goal modeling counterintuitive in a pervasive environment. Would not make it more sense to define the goal of a user in terms of rules? For example, a user prefer to specify that *if* some intrusion takes place, *then* the police must be alerted. Observe that this is allowed in WSML by means of Assumptions, but a discussion about goal specification is missing. This may be another genuine problem arising in this domain, i.e., finding optimal workflows to go from a set of specified Assumptions to a set of specified Effects, where these specifications are given at a conceptual level by a user and it is left to the system to look into the details of available services.
WSML vs OWL-S. As a *personal* opinion (based on previous work conducted with WSML and OWL-S), I find that the rule-oriented approach of WSML is indeed quite intuitive and easier to work with than OWL-S as the authors suggest. This may help to build simplified interfaces to let users modeling services, goals and rules. Otherwise, RDFS and OWL have now become quite popular and several ontologies are available (e.g., for geo-location, time and so on). Building a solution on top of WSML rather than OWL would benefit from having a strategy to guarantee backward (and possibly forward) compatibility with the RDF-based data. I think that this should be acknowledged in the comparison between OWL-S and WSML.
WSML constructs. I always found a bit overcomplicated the distinction between Preconditions, Assumptions, Post-conditions and Effects (reflected also in the quasi-similar definitions in Page 3). It would be interesting, while extending WSML, to take position on this distinction (is it needed or can be dropped for a more simple distinction between preconditions and effects?). Observe that in practice you have used Preconditions and Post-conditions for services and Effects for goals (why?).
The paper is reasonably well written and easy to read. However, the paper contains some typos (see detailed comments below).
One main obstacle to reading the paper is that the paper is not self-contained. Not everybody in the semantic web / AI / database community is familiar with Z-language. Although it may not be needed to formally introduce the language, every expression pattern used in the axioms (many axioms use a same pattern) should be at least explained in the text.
**Summary of the review
If the authors complete their research and provide substantial empirical evidence of the effectiveness of the proposed language and approach in solving a real-world interoperability problem, this can be an interesting contribution to the research field (and open a new line of work in this direction). However, in its current stage, the research seems too preliminary and the results not conclusive. I also suggest providing a comparison between the proposed solution and more recent models proposed in the SWS field.
there are a large number → there is a large number
by the Web → on the web
SWS is introduced (numbered item 2) before the acronym is resolved (numbered item 5)
“An interesting approach that enhances interoperabil- ity and scalability is to use some concepts adopted for services. “ → I suggest rephrasing this sentence. Suggested
adopting semantic description → adopting semantic descriptions
Figure 2 introduce the +is_a relation looping over pervasiveWS. In the listings I find “subServiceOf” which I cannot find in Figure 2. Are these ones the same relation? Otherwise, in the text, you use the word supertype, and the relation hasSuperWS.
Why instancePervasiveWS is not subclass of Service?
Explain one example for each axiom pattern used in Figure 3
Fig 16: the nextTo predicate is used. How to make this predicate interoperable with geolocalization devices is not explained.
“is a struct composed” → is a structure (?) composed
an other task → another task
pervasive environment designs → pervasive environment design
Marco-Ruiz, L., Pedrinaci, C., Maldonado, J., Panziera, L., Chen, R. and Bellika, J. (2016) Publication, Discovery and Interoperability of Clinical Decision Support Systems: a Linked Data Approach, Journal of Biomedical Informatics
Roman, D., Kopecky, J., Vitvar, T., Domingue, J. and Fensel, D. (2015) WSMO-Lite and hRESTS: Lightweight Semantic Annotations for Web Services and RESTful APIs, Web Semantics: Science, Services and Agents on the World Wide Web, Elsevier
Álvaro Villalba, Juan Luis Pérez, David Carrera, Carlos Pedrinaci, Luca Panziera:
servIoTicy and iServe: A Scalable Platform for Mining the IoT. ANT/SEIT 2015: 1022-1027
Pedrinaci, C., Cardoso, J. and Leidig, T. (2014) Linked USDL: a Vocabulary for Web-scale Service Trading, 11th Extended Semantic Web Conference (ESWC 2014), Springer
Luca Panziera, Marco Comerio, Matteo Palmonari, Flavio De Paoli, Carlo Batini:
Quality-driven Extraction, Fusion and Matchmaking of Semantic Web API Descriptions. J. Web Eng. 11(3): 247-268 (2012)
Mohsen Taheriyan, Craig A. Knoblock, Pedro A. Szekely, José Luis Ambite:
Rapidly Integrating Services into the Linked Data Cloud. International Semantic Web Conference (1) 2012: 559-574
Kopecky, J., Vitvar, T., Pedrinaci, C. and Maleshkova, M. (2011) RESTful Services with Lightweight Machine-readable Descriptions and Semantic Annotations, in eds. Erik Wilde,Cesare Pautasso, REST: From Research to Practice, Springer
Liu, D., Li, N., Pedrinaci, C., Kopecky, J., Maleshkova, M. and Domingue, J. (2011) An Approach to Construct Dynamic Service Mashups using Lightweight Semantics, Workshop: The 3rd International Workshop on Lightweight Integration on the Web (ComposableWeb 2011) at The 11th International Conference on Web Engineering (ICWE 2011)