User Profile Ontology for the Internet of Things

Tracking #: 2215-3428

Authors: 
Andrea Pabon-Guerrero
Lider Rojas
Miguel Niño

Responsible editor: 
Armin Haller

Submission type: 
Ontology Description
Abstract: 
Determining the context in which the user is located is key to personalizing services in the Internet of Things. A generic, semantic, interoperable mechanism that manages user profile is thus required. Using ontologies that can model user preferences and their context according to both their location and the activities they are executing, a user profile management mechanism for IoT is proposed that enables users to establish preferences on smart objects that when taken into account create personalized services with minimum human intervention. The methodology used in creating this user profile ontology facili-tates rapid building of a formal, evolutionary model. To validate the solution, a case study adapted a semantic interaction sce-nario of smart objects, in which incorporated real users measured the ability of these to personalize the services provided. The results suggest that the ontology offers a good solution for semantic interoperability scenarios of smart objects that adapt their services to user characteristics. Testing verified that services were adjusted automatically within an acceptable time (52 sec-onds) in comparison with manual adjustments that might require several minutes. The created ontology is suited to environ-ments of semantic interaction of smart objects, where its role is adapting services automatically, according to user context.
Full PDF Version: 
Tags: 
Reviewed

Decision/Status: 
Reject

Solicited Reviews:
Click to Expand/Collapse
Review #1
Anonymous submitted on 26/Jun/2019
Suggestion:
Reject
Review Comment:

This manuscript was submitted as 'Ontology Description' and should be reviewed along the following dimensions:
(1) Quality and relevance of the described ontology (convincing evidence must be provided).

- The authors did not provide convincing evidence about the novelty of their work. It is very hard to understand whether the developed ontology is an automatically developed ontology or a manually developed ontology. If it is an automatically developed ontology, the authors should compare it to a gold standard ontology by calculating the similarity between the graphical representation of two ontologies. It is possible to find numerous studies on this subject if the authors search for "ontology learning". They can also use a text corpus to see if the ontology can sufficiently represent the corpus. There are some other alternative ways to evaluate it as well.

If it is a manually developed ontology, they can still evaluate it on a text corpus or using different approaches. However, it has to be very clear that it has been manually developed.

- The authors claim that they have extended an ontology called, Semantic Object Ontology (SOO), which I have not been successful in locating it on the internet. When I looked at the presented ontology graph, it is not clear which concepts are created by the authors.

- There is an already available W3C ontology on a related subject. Although it does not solve the problem the authors somehow aim to tackle, it may be worth looking into and citing the SSN Ontology and the SOSA Ontology. It may enrich their related work section and may give them some ideas (see https://www.w3.org/TR/vocab-ssn/).

- There are also some issues in the evaluation section. First, the authors did not present any results in the evaluation section. It is full of description of the metrics used in the analysis. Second, the authors never mentioned which database technology they have used in their experiments. I would be very much interested to know this particular information as it can affect the performance of the IoT system a lot. Third, it is very difficult to picture a scenario where a resource constraint device, a sensor will download an ontology and annotate the data on the sensor. This is neither possible nor ideal in IoT systems. The usual practice is to send the data to a cluster head or a more powerful machine in order to transmit it to the cloud. The semantic annotations are usually done in the cloud or on the most powerful machine as the semantics come with its computational expense.

- Although the authors might have a novel idea, it has not been reflected in the paper. The authors need to revisit and reformulate their hypothesis, conduct new experiments, and submit a new manuscript to demonstrate their novel ideas.

(2) Illustration, clarity and readability of the describing paper, which shall convey to the reader the key aspects of the described ontology.

It is not written in a clear and readable way. It is difficult to follow and understand the novelty of the paper. The paper contradicted its novelties a few times. There are many typos not only throughout the paper but also in the developed ontology.

Review #2
Anonymous submitted on 28/Jun/2019
Suggestion:
Reject
Review Comment:

This manuscript was submitted as 'Ontology Description' and should be reviewed along the following dimensions: (1) Quality and relevance of the described ontology (convincing evidence must be provided). (2) Illustration, clarity and readability of the describing paper, which shall convey to the reader the key aspects of the described ontology.

Authors of this paper propose a user’ profile ontology for the internet of things. They evaluate the ontology through a user experience questionnaire of an application using the ontology.
Authors do not provide the ontology or (if the ontology is published) any link to the ontology. Therefore, it is difficult to evaluate the quality of the ontology.
Regarding the relevance of the ontology, authors mention that there is not a complete ontology with all the concepts they need. However, (as it is clear from table 2) a combination of only 2 existing ontologies will suffice for the authors’ needs. In fact, authors in section 4.1, 4.2 and 4.3 describe mostly concepts of existing ontologies.
Regarding the illustration and clarity of the paper, there are some inconsistencies in the figures. In figure 2, there are concepts (PUG:Activity) represented twice in the ontology graph. In Figures 3, 4 and 5 some prefixes have disappeared. Therefore, it is not clear whether authors have linked/aligned concepts with existing ontologies or they simply borrow the ideas. Regarding the readability in general the paper is easy to read, however there are parts difficult to understand throughout the text. In particular, the abstract repeats the same ideas more than once and have long sentences difficult to follow.

I also missed some well-known references in the field of semantics (ontologies) in general, and semantics in the fields of IoT and WoT in particular. Especially, in the first paragraphs, where authors talk about general semantics and WoT.

Authors have created a system which makes use of the ontology. The evaluation in the paper is based on user experience questionnaires about the system (no the ontology). However, the description of the system is quite vague, and therefore it is difficult to know if the evaluation is sound. In any case, from the description of the system and evaluation, it is impossible to reproduce it. For example, it is difficult to understand why in the first “iteration” not all the services where configured. It is not clear also whether there was an improvement in the system between both “iterations”; authors need to detail the improvements performed. Also, the sample size is quite small. Regarding this system, authors mention one limitation as the use of the system by only one person at a time (with one user profile); I think this is a huge limitation, that need to be solved, and I do not see a priori the difficulty of solving it. Therefore, I advise authors to solve it or to explain the difficulties found in the solution. Authors mention that the system learns the recurring behaviors, but authors need to explain the techniques and algorithms used to this end.

Therefore, I see that the paper in its current state is poor as an ontology description and poor as a system implementation. Therefore, I encourage authors to focus on one of the two and produce a more detailed and reproducible paper, as I think the system has some potential and it could be useful to end users.

Additionally, to mention some minor details; authors mention “SPARQL queries as a reasoning method”; I would not classify SPARQL as a reasoning method per se. There is a reference in table 2 which is not in table 1. The “knowledges” concept in Figure 1 and in the text is confusing, authors need to define this concept in the text. The references to Methondology and CASE tool are missed. As a future work authors mention the use of “different” reasoning techniques, however, they have not described any reasoning technique at all in the paper.

There are some concepts that need a description, for example I have some doubts about the meaning of “Characteristics” (section 4.1). And I have some doubts about the differences between the concepts “Things” and “Objects”.

Review #3
By Nicholas Car submitted on 03/Oct/2019
Suggestion:
Reject
Review Comment:

(1) Quality and relevance of the described ontology (convincing evidence must be provided).

This paper does not cite well-known Semantic Web IoT work, in particular the Semantic Sensor Networks Ontology which both appears in recent editions of this journal [1] and is a W3C Recommendation [2]. The authors do not have to use SSN but they do have to cite it and, at least, state how their ontology differs from it in its conceptualisation of sensors etc.

The paper claims "this project also implements a strategy to bring data related to the user from social networks,in the particular case from Google, in order to facilitate the collection of user data with minimal intervention" but this is not explained. Either it must be explained/proved or the claim removed.

The examples in Table 4 Ontology Assertions are all datatype properties (with literal values). Firstly, some do not follow OWL datatype best practice (all are xsd:string, some should be xsd:date, xsd:integer etc). Secondly, it is not clear to me how preferences are understood by the system other than as a named property with a range (min/max). How would preferences with boolean or list selection values be expressed? Thirdly, it is not clear to me how Preferences/Interests etc. are compared. What if two Preferences collide in system understood instructions? Is the "Score_preference" used somehow to determine precedence?

No logic is shown as to how the system can compare or scale preference values. A claim of the paper is that "Performing reasoning on the ontology, it was possible to configure the services that smart objects provide automatically or with minimal user intervention." with the scenario being that a user "comes into the room every day between 7 and 8 pm, always turning on the bulb via CMA. After a few days, the object registers this as a recurring behavior". It is not clear to me how the systems can recognise patterns such as this example of recurring behaviour and record this as a user preference. There could be many ways to interpret actions other than by their temporal repetition, e.g where the user is so how does the system know that time of day is the important property here? Perhaps only temporal patterning is sought? How would any events *other than* clock time be indicated? There must be lots more logic/software not mentioned to do this. Either this logic must be given or the claim removed.

[1] http://www.semantic-web-journal.net/content/modular-ssn-ontology-joint-w...
[2] https://www.w3.org/TR/vocab-ssn/

(2) Illustration, clarity and readability of the describing paper, which shall convey to the reader the key aspects of the described ontology.

Web of Things (WoT) is used once only (first line of Introduction). Replace with IoT.

The ontology appears in informal diagrammatic form only, allbethey auto-generated diagrams from code: no formal diagramming is used (like UML), no machine-readable ontology specification itself is given (code) and no links to code elsewhere either. This makes it impossible to review the elements/axioms of the ontology directly. Either formal diagramming, code or links to it must be given.

There are inconsistent uses of prefixes in the ontology diagrams. E.g. Fig. 2 "PUG:Person", Fig. 3 "Person". While I can work some of this out (I assume they are the same) they need to be aligned.

Many spelling mistakes, some in tabular info e.g. Table 4 "Name_interes", should be "Name_interest". There are numerous other small punctuation errors too (e.g. italicise _et al._), too many to list individually here.

OVERALL
The very interesting claims in this paper, that user preferences can either be entered or, even more interestingly, learned by the system, and not explained. How does a user enter preference the system an understand? How can the system learn preferences it can understand from behaviour? Are there a set of fixed preference types the system can understand or is this open-ended? You interest/preferences modelling is taken from Golemati and Golemati uses an "Interest hierarchy" to order preferences but you don't mention anything about such a breakdown here. Golemati also gives discrete values for properties, such as User Context that are easy to trigger systems responses to but you don't mention anything similar here: the only trigger you give is temporal.

The ontology, as documented here, seems to have far too much detail in some areas (Fig.1 contains lots of building type information) and not enough in others (the makeup of preferences) to convince me that it actually works.

Your testing shows me something but I can't really work out if it's showing me that the system is actually responding to particular preferences.

At this stage, I'm not convinced that the ontology is capturing enough information to be able to do the job you claim it can or that the system can interpret what is captured usefully.

I think your research is interesting in concept but more proof of the ontology working is needed or you have to instead qualify why you've designed the ontology in a particular way, particularly around preference expression, but then not include experiments that don't relate to that design.