Semantic Web Technologies for the Internet of Things

ContentslistsavailableatScienceDirect

Internet of Things

journalhomepage:www.elsevier.com/locate/iot

Semantic Web Technologies for the Internet of Things:

Systematic Literature Review

Ahlem Rhayem

^∗

, Mohamed Ben Ahmed Mhiri, Faiez Gargouri

Miracl-laboratory, Sfax University, Tunisia

a r t i c l e i n f o

Article history:

Received 5 February 2020 Revised 27 March 2020 Accepted 28 April 2020 Available online 6 May 2020 Keywords:

Internet of Things Web of Things

Semantic Web Technologies Ontology

Interoperability Standards

a b s t r a c t

Nowadays,the useof theInternet ofThings(IoT)in diverseapplications becomesvery popular.Accordingly,aproliferation of objectswith remote sensing, actuation,analysis, andsharing capabilitieswillbe interconnectedontopofheterogeneouscommunication networks. Theirdeployment contexts arecontinuously changed, whichimply achange intheirdescriptionsandcharacteristics.Inaddition,theyareafundamental sourceofa hugequantityofgathereddatawithdifferentencoding formats.Accordingly,thisdatais badlyexpressed, understoodand exploitedby othersystemsand devices.Fromthisre- gard,severalchallengesassociated withstandardization,interoperability, discovery,secu- rity,anddescriptionofIoTresourcesand theircorrespondingdatahaveemerged.Inthis context,SemanticWebTechnologies(SWT)seemasuitableandanefficientsolutiontore- lievethesechallenges.Therefore,aSystematicLiteratureReview(SLR)methodologyisper- formedtoinvestiagteandanalyzeasetofthemostrecentandrelevantapproachesthat dealwithSWTintheIoTdomain.Theseapproachesarediscussedandevaluatedbasedon sevendifferentresearchquestions.Finally,future insightsand researchopportunities are suggested.

© 2020 Elsevier B.V. All rights reserved.

1. Introduction 1.1. Context

Inorderto ensure acomfortablelife, a myriadofstakeholders become interestedinintegratingdailylife objects into thenetworkinthefield ofinformationtechnologies. Sensors,actuators,andRFIDtags arewidely exploitedtoenablethis integration.Fromthisperspective,usershavetheabilitytoaccessreal-timeinformationgatheredbyconnectedobjectsany timeandanywhere.Theevolutionofusinginternetwithreal-worldobjectshasledtotheriseoftheInternetofThings(IoT).

The International TelecommunicationUnion defines the IoT as a global infrastructure forthe Information Society, which enablesadvancedservicestoconnect(physicalandvirtual)thingsbasedonexistingandevolvinginteroperableinformation andcommunicationtechnologies [1]. Inaddition, Perera etal.[2]affirmed that ”The IoT allows people and things to be connectedAnytime,Anyplace,withAnythingandAnyone,ideallyusingAnypath/networkandAnyservice”.Infact,billions ofdevicesandnode sensors,becomeconnectedtoeach otherwithpotential capabilitiestosense,communicateandshare dataandinformationabouttheirsurroundingenvironment.

∗ Corresponding author.

E-mail addresses: ahlemrhayem@gmail.com (A. Rhayem), med.mhiri@gmail.com (M.B.A. Mhiri), faiez.gargouri@isims.usf.tn (F. Gargouri).

https://doi.org/10.1016/j.iot.2020.100206

2542-6605/© 2020 Elsevier B.V. All rights reserved.

IoTismainlybasedontheWirelessSensorNetworksfield(WSN)thathasemergedasoneofthemostpromisingtechnol- ogy.Thankstotheintelligentdeployedsensors,WSNisgreatlyemployedinmanyapplicationdomainsthatensureefficient supervision.

IoThasrecognizedaregularevolutionbyexploitingitsservicesfromtheweb;knownastheWebofThings(WoT).

Forabetterexplanationofthefundamentalconceptsofthesetechnologies,aspecialfocusisputontheirsemantics.

1.2. Problemstatement

Despiteits importance, the continuousevolution ofthe IoT hasled to a complexitylevel dueto thehuge amount of heterogeneousdeployedobjects,sensingdata,andsuggestedservices.Ontheonehand,theconnectedobjectsaredesigned by differentmanufacturers(Google,IBM,Nokia,etc).Accordingly, thegenerateddatahasdifferentcoding formatsleading to a complex data exchange task.Moreover, this amoutof data,including semantic heterogeneity (synonymy, antonymy, polysemy,etc),keepsgrowing.Inthisrespect,theobtainedmeasurecanbeexpressedindifferentterms.Forinstance,tem- peraturecanbemeasuredwiththeCelsius,Fahrenheitorkelvindegree.Ontheotherhand,thevarietyoftheseobjectsand theircontinuouslychangingrequirementsanddeploymentcontextsfurthercomplicatetheirmanagementandconfiguration tasks.ThesechallengesareraisedduetotheabsenceofaunifiedandstandardmodelofIoTdevicesalongwiththeirdata and services.Thereby, the notionof semantics plays a key role inthe Internet of Thingsdue to its efficiencyin solving the problemsofheterogeneity,interoperability, anddatainterpretation.Furthermore, semanticinteroperabilitymeans the abilityofdifferentpartiestoaccessandinterpretunambiguousdatasincetheconnectedobjectsareabletoexchangedata witheachother andwithotherusers online.SWT exploitationintheIoT field providesan explicit,easyandcomprehen- sibledescriptioninordertoexpresssemanticobjectsandtheirdata.Inaddition,semanticssubscribestodefine consensus that facilitates datasetsharing,reuse,integration, andinterrogation. Thisdata,extractedfromdifferent connectedobjects anddomains,ensures cooperationbetweentheconnectedobjects byfacilitating thecommunicationbetweentheminor- der tohighlighttheirintelligenceaspects. Moreover,itensures dataanalysisandreasoningthat explaintheneed fordata alignmentwithdifferentvocabulariesandframeworks.

Forthispurpose,applyingSWTinIoTdomainisconsideredasasuitableandreasonablesolution.Thisaimwasconfirmed byBarnaghietal.[3]whoclaimedthatapplyingsemantictechnologiestoIoTpromotesinteroperabilityamongitsresources, informationmodels,dataprovidersandconsumers.Italsofacilitateseffectivedataaccessandintegration,resourcediscovery, semanticreasoningandknowledgeextraction.

Infact,the synergybetweenSWTandIoTorWoTdomains givesrisetothebirth ofanewappellation;known asthe semanticweb ofthings (SWoT).According to[4],[5]SWoT isan emergingvision, thatbrings together the semanticweb andtheInternetofThings.

1.3. Motivationandcontribution

Toensure a reliable semanticmodeling, ontologies are widely usedfor IoT devicesanddata annotation.According to Studer etal. [6], an ontology isa formal andexplicit specificationof a shared conceptualization.It is used to represent knowledgeinthesetofrelatedconceptfields.

Itmainlyconsistof:

• Definingstandardvocabulariesthatwillbesharedandreusedbetweenobjectsaswellasbetweenhumans.

• Facilitatingthediscovery,integration,manipulationandconfigurationofIoTresourcesandtheirdata.

• Supportingreasoningmechanismsforinferringintelligentdecisions.

According to Yeet al.[7],ontologies can be classified onthe grounds oftheir expressiveness(light-weight orheavy- weight)orgenerality(generic,domain-specific,orapplication-specific).Inthiswork,ourfocusisrestrictedtotheIoT/WoT domainingeneralwithouttakingintoaccountitsapplications.

Wedetail,classify, compareanddiscussthemostrecentrealizedsemanticmodelingapproachesintheIoT/WoTdomain througha taxonomyoftheprevious approaches.Webelievethat thissurvey willhelpbothresearchersanddevelopersto focusontheamalgamationbetweenSWTandtheIoT/WoTdomainthatwillguidethemtowardsfutureresearchdirections.

WebelievethatthissurveywillhelpbothresearchersanddeveloperstofocusontheamalgamationbetweenSWTand thesedomainsasitwillguidethemtowardsnewfutureresearchareas.

Theprimarycontributionsofourworkarelistedasfollows:

• Weprovideanextensivereviewoftheup-todateresearchprogressabouttheroleofSWoT.

• Weproposeanexhaustiveclassificationwhichcontributestorepresentingadeepanalysisofacomprehensiveliterature review.

• Weprovidein-depthcomparisonsanddiscussionsofvariousapproaches, whosemajorroleistoensuresemanticinter- operabilityinthecontextofIoT,WoT,andsensors.

• WeestablishsignificantfuturetrendsforsemanticsintheIoTdomain.

1.4.Paperorganization

Theorganizationofthispaperisasfollows:InSection2,weprovideadescriptionofrelatedsurveypapersforsemantics IoT.Section 3highlights themethodology ofoursurvey.InSection 4,weintendto providea breakdownofthereviewed papersforsemanticsrepresentationinthisfield. InSection 5, anextensivediscussion oftheobjectivesofthe studiedre- searchisprovided.Section6extensivelydescribespossiblefutureresearchopportunitiesbasedonthesystematicreviewin theprevioussections.Finally,conclusionswillbedrawninSection7.

2. Relatedsurveypapers

Oursurvey isrelatedto thestate-of-the-artdealingwith semanticsin theIoT, WoTandWSN domains leadingto the evaluationofsomeresearch survey papers.Thereby,inthissection,we state therecentproposed survey papersonthese topics.

Thefirstinvestigationinthisresearcharea wasrevealedin2012byBarnaghietal.[3]whohaveexplainedtheimpor- tanceofdefining andpresentingIoTsemantics inorderto resolvethe heterogeneityandambiguityof thehugecollected datathroughconnectedobjectsandtoensuretheinteroperabilitybetweenIoTsystems.Fromthisperspective,theyproposed anoverviewofsomeexistingontologiesthataimatrepresentingsensorsandtheirdata,suchasO&MandSSNontologies.

In2016,thesurveywork,realizedbySzilaggietal.in[8],highlightedanoverviewofSWTusedatdifferentIoTsystem layers andthewell-important ontologies that wereused to develop applicationsandservices inIoT, namelySSN [9],IoT ontology[10],andtheIoT-O[11]

In2017,Bajajaetal.[12]studied,discussedandanalyzedvariousontologiesthatwillbereusedintheIoTdomainbytak- ingintoaccountsensors,time,locationandcontext-awareness.Theaforementionedapproacheswereclassifiedintogeneric anddomain specific ontologies. In addition, the authors in [13], outlined the existing IoT ontologies to ensure semantic interoperabilitybetweenheterogeneousIoTsystems.TheymainlyfocusedongenericontologythatwasappliedinIoTplat- formsanddomain-specificapplicationontologies(healthcareandlogisticdomains).Deetal.[14]setforthasurveyofthe currentstate-of-the-artontheuseofontologiesintheWebofThings(WoT),whichrangedthemintwolayered-approaches.

Thecross-domainframeworkrepresentstheconceptsofWoTelements(devices,services,data,etc.)whilethedomainlayer statessomedevelopedontologiesthataregroupedintoenvironment(smarthome,agriculture,andsoon.)anduser-oriented (healthcare,e-learning,andothers)domains.

In2018,Androecetal.[15]collectedandclassifieddiverseworkstoensure semanticinteroperabilityinIoT.Theycon- ducteda systematicliteraturereview methodologyin their survey.Theaforementioned studiesare arranged accordingto theiryearofpublication,country,andmaincontribution.ThisstudycoversdifferentIoTapplicationdomains,namelyhealth- care,smartcity,etc. Inthiswork, theauthorslightlyintroducetheselected approaches.Therefore,severalrecentandim- portantstudiesarenotconsideredinthissurvey.

Tables1and2recapitulateanddiscussthesurveysandoverviewsrelatedtoSWoTbyhighlightingtheirprincipalassets anddrawbacks.

Despitetheirencouragingandneatoutcomes,theaforementionedeffortsshowedremarkableshortcomingsinthecom- plementarityandthebreadthofunderstandingsemanticsinWSN,IoTandWoTdomains.

Ourworkisdistinguishedby:

• Theprimarygoalofthissurveyistoconsiderandexplainmorerecentandcitedworksthanthosementionedinprevious surveys.

• We coverdiverseIoT semanticlevels(resource,data,service, security).Tothebestofourknowledge,noneofthepro- posedsurveyshavecoveredtheseaspects.

• We propose an indepth-analysisof thehighlighted approachesin whichwe classify them intoseveralcategoriesde- scribingtheirmainfeatures.Thenwesuggestacomparisonandadiscussionoftheoutlinedapproaches.Inaddition,we suggestseveralopenresearchquestionsforreaders.

• Thissurveypaperconsistsofanewclassificationofrelatedworks.

• We offer a set of research questions to evaluate and compare related worksin order to understand their shortfalls, strengths,andchallenges.

• WediscussfuturedirectionsofIoTdomain,whicharesketchelyaddressedwithintheIoTdomainintheexistingsurveys.

3. Surveymethodology

Inthispaper,weconcentratedonasystematicliteraturereview(SLR)[16,17]toanalyzeandevaluaterelevantcontribu- tionsrelatedto SWoT.As elucidatedby [16],SLRiscomposedofthree mainphases:planningthereview,performing the review,andreportingthereview.Thesephaseswillbedescribedbelow.Fig.1showstheadoptedprocedure.

3.1. Planningreview:Goalsandresearchquestions

Thisphaseconsistsindevelopingaprotocolreviewinordertodrawthemainobjectivesofthissystematicreview,outline theadoptedresearchquestionsandproposetheresearchstrategy.

Table 1

Summary of IoT related surveys.

Items Year Assets Drawbacks Application

Domain

Adopted Methodology [3] 2012 -This work defined the semantic levels

that should be considered in the IoT domain (resource level, data level, service level, security level and application level) to ensure interoperability between IoT systems.

-It gave an overview of important works that were previously suggested in order to define IoT semantics, such as SSN and O& M ontologies

-Initial work has focused only on approaches that were suggested before and during 2012, like such as SSN and O&M.

-This work has not made a comparison to its related works in terms of their main contributions, methodologies, features, etc.

IoT ×

[8] 2016 - The authors in this work explained the importance of using SWTs (OWL, RDF, DL) at different levels of IoT systems to guarantee semantic interoperability between these heterogeneous systems.

- It gave an overview of the most cited ontologies that were proposed for IoT knowledge representation, such as SSN, SAN, and IoT-O.

- -Diverse relevant approaches were not included in this survey. The latter highlights only the works that were conducted before and during 2016.

- The authors did not follow a methodology to select the cited approaches. Hence, there is neither a comparison nor a classification in this work.

IoT ×

[12] 2017 - This paper provided an overview of the most important ontologies that should be considered in the IoT domain such as sensor ontologies, time ontologies, location ontologies and so on.

- It classified the existing approaches into generic and domain-specific categories.

- A weak comparison of the previous research works that did not consider semantic features to build their ontologies; including semantic language, and deployment method (modular, monolithic, from scratch, etc).

- Limited semantic IoT approaches are presented in this paper. Approaches related to security aspects and WoT are not mentioned.

- Absence of a notable discussion concerning the drawbacks of the studied works.

IoT ×

[13] 2017 - This work described related studies whose major role consists in defining ontologies not only in the IoT domain but also in healthcare and logistic applications in order to achieve semantic interoperability between IoT systems.

- The studied approaches in the IoT domain emphasized only domain-specific aspects of sensors and sensor networks. Approaches about IoT services, security and WoT are missing.

- This work’s approaches are just mentioned without comparing their semantic features (main scope, methodology, ontology reuse, semantic web language, etc) or discussing their drawbacks

IoT, Health care, Trans- portation and Logistics

×

Table 2

Summary of IoT related surveys (continued) Table 1.

Items Year Assets Drawbacks Application

Domain

Adopted Methodology [14] 2017 This work presented an overview of

the proposed ontologies in the IoT/WoT domain and classified these approaches into generic and domain levels.

- The studied approaches are sensor ontologies, location ontologies, service ontologies, data ontologies and other ontologies for a specific application (home, healthcare, etc).

- This paper focused more on the existing ontologies that can be reused in WoT (sensor ontologies, data ontologies, service ontologies, etc) without taking into consideration other domain-specific aspects, such as IoT-lite and IoT-O.

- Various IoT aspects including security and WoT are not addressed.

WoT ×

[15] 2018 This work proposed a systematic literature review of SWT application in the IoT domain.

The cited works are discussed and classified according to their publication dates, publication types (conference, journal, thesis, etc.), semantic usages,and citations in others works.

They do not compare the semantic features (semantic languages, methodologies, evaluations,

modularizations, etc) of the cited approaches.

Diverse IoT aspects are not highlighted in this survey, such as security, IoT service description, discovery, and WoT.

IoT Systematic

Literature Review

Fig. 1. Systematic literature review applied steps adapted from [16] .

ThepresentsurveypaperfocusesoninvestigatingandexploringtheamalgamationofSWTandIoT.Section2summarizes thepreviousproposedsurveysandhighlightsthecontributionsofourworkcomparedtothepreviousones.Thereby,inour paper,weanalyzeanddiscusstherecentreviewedworksthroughasetofaccurateresearchquestions.Wemainlyfocuson:

• How can the semantic interoperability between smart objects, data, services, and applications in the IoT domain be assessed?

Guidedbythismainresearchquestion,aspecialheedispaidtothefollowingresearchquestions:

• RQ1:WhatarethemainobjectivesofapplyingSWTintheIoTdomain?

• RQ2:WhichaspectsoftheSWTstackareused?

• RQ3:Aretheproposedontologiesbasedonotherexistingontologiesoraretheybuiltfromscratch?

• RQ4:Didtheproposedapproachpresentamodularizedrepresentation?

• RQ5:Whichcontextsconcerningdeploymentdevicesareconsideredintheproposedmodels?

• RQ6:Whichmethodologydotheauthorsfollowtobuildtheirmodels?

• RQ7:Howdotheauthorsevaluatetheirmodels?

3.2.Conductingthereview

Thisphasefocusesonidentifying andselectingrelevantpapersthat can providearesponse toourresearch questions.

Thisstepismainlymadeupofaresearchstrategyandstudyselectionsteps.

Fig. 2. Iterative Papers selection results after applying inclusion and exclusion criteria.

3.2.1. Researchstrategy

Inour survey,we useWeb ofScience,Google Scholar,IEEEXploreDigital Library,ElsevierScopus, ACMdigitallibrary, Citeseerlibrary,ScienceDirect,andarXiv.orginordertosearchforrelevantscientificcontributions.Inaddition,weconsider thelinkedopenvocabulariesfortheIoT¹,IoTschema² andWebofThingsWorkingGroup³ thatareinterestedininteroper- abilityissueintheIoTdomain.Infact,welimitourkeywordstoInternetofthings,WebofThings,Sensors,Ontologiesand Semantics.Consequently,weusethefollowingresearchstring:((SensorORWSNORInternetofThingsORWebofThings) AND(SemanticWebTechnologiesORLogicsORontology)).

3.2.2. Studyselection

After searchingpublicationsinonlinedatabases andfilteringandscreeningthereturned papers,an initialset of1400 studies wasfound. The latter containsnot only studies that do not focuson the specified research problem(e.g. survey papersaboutIoT)butalsoduplicatepapers(e.g.appearedindifferentdatabases,PhDthesesandrelatedpapers,duplicate languages etc.)that are proposed fordifferentapplication domains along withothers.Wherefore,a selection step ofthe relevantpapersbecomesnecessarybasedonseveralinclusionandexclusioncriteria.

Thelatteraredefinedduringthisphasetoselectthemostrecentandrelevantworks.First,weonlyconsideredthecon- tributionsthatwerepublishedinpeer-reviewedjournals,conferences,andbookchaptersfrom2011to2020.So,thenumber ofpaperswasreducedfrom1400to762.Second,weexaminedworkswhosemajorfocuswasonIoT,sensornetworksand WoTingeneralwithouttakingintoconsiderationIoT domainapplications,likeHealthcare, logistics,industries,andsoon.

Consequently,weobtained442papers.Third,thenumberofpaperswasreducedtojust198 afterremoving duplicatepa- pers,Ph.D.theses,surveypapers,shortpapers,andnon-Englishpapers.Afterthat,wehaveinvestigatedtheimportanceof thesepapers basedon their titles, abstracts,andkeywords. Accordingly, froma setof75 selectedpapers that were fully read,wefinallyfocused on37 relevantstudiesinthiswork.Eachcontributionisanalyzedtoextractpertinentinformation abouttheproposedresearchquestions.Fig.2showstheprocess.

3.3. Reportingthereview

Thisphaseismainlycomprises threemainsteps,suchassummarizing, comparing,anddiscussingtheselectedpapers.

Thesestepsaredescribedbelow.

3.3.1. Summarizingselectedpapers

IoTsemanticscanbedefinedinalayeredarchitecture[3]asdescribedinFig.3.ThefirstlayerrepresentstheIoTresource thataimstodefinesemanticreal-worldobjectsandnetworks. Thesecondoneis,however,fordatarepresentationinorder todescribeIoTresourcesandtheirsemanticdataandexplainthewaySWTexploitsthisrepresentationformanagementand interpretationgoals.Theserviceapplication, atthethirdlayer,revealshowSWThelpsdevelopIoTapplicationsandrecom- mendadequateservicesforusers.Thesecuritylayershowsthecapabilitiesofthesetechnologiestorepresentvulnerabilities andsecuritymechanismsintheIoTdomain.Inspiredbythisarchitectureandinordertoanswerthefirstresearchquestion,

”WhatarethemainobjectivesofapplyingSWTintheIoT?”,weclassifythestudiedapproachesintofiveIoTrelatedaspects, accordingtotheir maincontributionnamelyIoTdata,IoT dataandservice,IoTservice,IoT securityandWoT.Infact,Fig.4 illustratesthisclassificationwheretheIoTresourcesarealreadyconsideredbyotherIoTaspectsasfollows:

1LOV4IoT: https://sensormeasurement.appspot.com/

2iot.schema.org

3https://www.w3.org/WoT/WG/

Fig. 3. Semantic layer in Internet of Things, adapted from [3] .

Fig. 4. Classification of IoT-based Semantic Approaches.

• IoT data:Thiscategorysummarizesthe researchworksthatproposed asemanticmodelofIoT resourcesandtheirde- tecteddata.

• IoTservices:Atthislevel,theproposedworksemphasizetheservicesgivenbyIoTresources.

• IoTdata&services:Thiscategorydescribesthesuggestedworksthatdefineasemanticmodelforbothdataandservices generatedbyIoTresources.

• IoTsecuritydimension:identifiesapproachesthatareofferedtoensureIoTsecurity.

• WoTdimension: atthis level,aseamless interconnection betweentheWorld Wide Web(WWW) andIoT resources is establishedinordertodefinetheWebofThings.Consequently,heterogeneousdevicescanbeeasilyusedandconfigured throughawebapplication.Thus,thiscanensureefficientcommunication(monitoring,integration,deployment,deletion, etc.)betweenusers anddevicesby usingweb standards(HTTP/WebSockets). Fromthis respect,recentresearch works havedeeplyfocusedonSWTdeploymentinWoT,whichwillbeattheheartofourwork.

3.3.2. Comparingreviewedapproaches

Afterreviewingeachapproach,weproposeacomparativestudyaccordingtoseveralcriteria.Thesecriteriaareexplained asfollows:

• SemanticTechnologies:Theyareproposedtogiveananswertothisquestion”WhichaspectsoftheSWTstackareused”.

This dimension specifies themodeling, reasoning andinterrogating language (OWL, RDF,SWRL, SPARQL, etc.). In fact, ontologiesaredevelopedbasedonsemanticweblanguagesthatofferformalsemanticsandaprecisesyntacticstructure ofthedomainknowledge.Inwhatfollows,wewillpayspecialattentiontothemostknownlanguages.

TheResource Description Framework(RDF): According to the W3C⁴, RDF isa graphthat definesa standard model for datainterchangeontheweb.ItisalsocalledanRDFtripleasitcomprisesasubject, object,andpredicate.Thesubject defines the semanticentity of interest, the object refers to the value of this entityand the predicaterepresents the relationbetweentheentityandits value.RDF Schema(RDFS)is anextension ofRDF thatprovides a structureddata- modelingvocabularyforRDFdata.Itdefinesgeneralizationrelations(rdfs;Class,rdfs:subPropertyOf,etc)andconstraints (rdfs;domain,rdfs:range)ofRDFtriples.

TheOntologyWeb Language (OWL):is an ontology language that isstandardized by theW3Cfor thesemanticweb. It allowstherepresentationofconceptsandtheirrelationshipwithlogicalreasoningandinformationsystemsinanappro- priatemanner.Itiseasierandmoreexpressive comparedtotheRDFSlanguage.In2009,theW3COWLworkinggroup hasproposedanextension ofOWL, knownasOWL2⁵.OWL2hasthreedifferentprofiles,namelyOWL2EL,OWL2QL, andOWL2RL.Thefirstprofileisemployedforontologiesthatcontainalargenumberofclassesandproperties.Thesec- ondprofileisveryusefulforalargeontologythatincludesamassivequantityofinstances.Thethirdprofileisapplicable toapplicationsthatrequirescalablereasoning,intimethatispolynomialwithrespecttothesizeoftheontology.

TheSemanticWebRuleLanguage (SWRL):isarulelanguage forthesemanticweb. Itallowsuserstodefine rulesbased on OWLconcepts and their relationshipsin orderto infer hiddenknowledge. It can be extended by usingfunctions;

namedbuilt-insthatcanbe mathematicaloperations, stringoperations,dates,andsoon.Furthermore,itpermitsusers todefinetheirownbuilt-insthatrespondtotheirrequirements.TheSWRLlanguageiscomposedoftwomainparts.The antecedent partrefers toa setofconditionsthat should be verifiedwhile theconsequencepartdescribesresults and actionsthatwillbeexecuted.

TheShapeConstraintLanguage(SHACL)⁶:isaW3CspecificationsthatfocusedonthevalidationofRDFgraphviaasetof conditionsandconstraints.Theseconditionsareprovidedasshapes.

TheSPARQLQueryLanguage:AccordingtotheW3C,SPARQLisaquerylanguagefortheRDFgraph.Ithelpssearch,add, remove,andupdateRDFdata.Italsosupportsextensiblevaluetestingandconstraining.TheresultofaSPARQLqueryis asetofRDFgraphs.

• Reuseofontologies:This criterionspecifies iftheauthorstakinginto accountthe previousexisting ontologies torep- resent the IoT domain knowledge,or ratherdefine their ontology fromscratch. This criterion givesan answerto the question”Aretheproposedontologiesbasedonotherexistingontologiesoraretheybuiltfromscratch?”.Infact,during theontologydevelopmentprocess,it’sveryimportanttoreuseexisting ontologiesratherthanbuild itfromscratch for manyreasons.Meanwhile,accordingtoLonsdale etal.[18],reusingan existingontology guaranteesthe qualityofthe newontology becausethe reused concepts have alreadybeen tested andvalidated.In addition, buildingan ontology fromscratchiscostlyandneedsgreathumaneffortsthatwillbe reducedbyreusingexistingones.Moreover,mapping betweentwoontologies,thatsharecomponentsthroughontologyreuse,iseasier.

• Modularity:Itverifies iftheproposed models arepresented bymodules ornot.This criteriongives ananswer tothe question”Didtheproposedapproachpresentamodularizedrepresentation?”.Infact,amonolithicontologyisverydiffi- culttobehandledandreused[19],[20],[21].Inaddition,ascalabilityproblemcanberevealedbymonolithicontology- basedapplications.Thisisjustifiedbytheamountofdatathatmustbeanalyzedbysemanticwebtechnologies(SWRL, SPARQL,RDF,).Inthisregard,modularontologyisa promisingsolutionasitcanfacilitateknowledgereuseacrosshet- erogeneousdomains.Itiseasiertomaintain,reuse,andmanagedistributedengineeringofontologymodulesindifferent locationsandareasofexpertise[22].Accordingly,modularityisanimportantchoiceinIoTforontologydevelopment,as asetofinterlinkedmodules,inordertoimprovescalability,reasoningperformanceandreusabledomainconceptsbased onnecessarymodules.

• Context:Thiscriterion answerstheresearch questionRQ5:”Whichcontextconcerning deploymentdevicesareconsid- eredintheproposedmodels?”.Infact,withitsswiftgrowth,thenetworkeddeploymentofobjectsbecomemorecom- plicatedanddifficulttobehandledthanever.ItreferstoIoTresourcecontextsthat shouldbetakenintoconsideration toensureanefficientandpermanentconfigurationandmanagementoftheseresources.

Deyin [23]hasdefinedcontext as”anyinformationthatcanbe usedto characterizethesituationofan entity.An entity isauser,a place,oraphysicalorcomputationalobjectthatisconsideredrelevanttotheinteractionbetweenauserandan application,includingtheuserandapplicationthemselves”.

TheconnectedobjectconstitutesthemainentityoftheIoTfield.Therefore,weelucidatefivecontextsthatarerelativeto objectdeployment, namelyinterconnectivitycontext(CoI),time context(CoT),locationcontext(CoL),trajectorycontext (CoTr),andobject’srequirementcontext(CoR).

Timecontext(CoT):Withthecontinuousprogressofnetworkedobjects,temporalissueshavebeenmorepowerfulthan before.Timecontextisemployedtoexplainthetemporalmodelingandreasoningfeaturesofconnectedthings.Accord- ingly,itimprovesthequalityofservicesprovidedbytheseobjects.

Locationcontext(CoL):Locationawarenessisinextricablylinkedtoconnectedobjects.Itisstronglyimportanttoobtain andrepresentinformationabouttheirpositioninthephysicalenvironment.

4https://www.w3.org/

5https://www.w3.org/TR/owl2-overview/

6https://www.w3.org/TR/shacl/

Trajectorycontext(CoTr):Context-awaretrajectoryrepresentsthemobilitycharacteristicsofconnectedobjects.Itisbased onthetwoabove-mentionedcontexts(timeandlocation).Thereby,itreferstoalocationlistthatiscrossedbyanobject duringapredefinedperiodoftime.Nonetheless,thereisalackofapplyingSWTtorepresentthetrajectorybehaviorsof connectedobjects.

Inter-connectivitycontext(CoI):Itisthefactofhavingknowledgeabouttheusedtechnologiestointerconnectreal-world objectstoInternet.Infact,largeamountsofshareddatathatconsumealotofnetworkbandwidtharegenerated.There- fore,thenetworkhastobeeffectivelymanagedforefficientutilization.

Object’srequirementcontext(CoR):thiscontextfocusesontheobject’sspecifications,suchasthebatterylevel,memory capacity,coveragerange,lifetime,andsoon.Theseinformationareextremelysalientduringtheirstateconfiguration.

• Methodology: Thiscriterionhelpsusidentifywhichmethodology isfollowedtobuild ontologiesinIoT.Itisrelatedto theresearch question”RQ6:Whichmethodologydotheauthorsfollowtobuildtheir models?”.Infact,anontology isa promisingsolutiontoensuresemanticinteroperabilitybetweenheterogeneoussystems.Buildingan IoTontologyisnot astraightforwardtaskduetothecomplexityoftheIoTdomainintermsoftheexponentialgrowthofdeployeddevices, continuousdeploymentcontexts,heterogeneousdata,etc.ThisneedscollaborationbetweentheIoTdomainandsoftware engineering experts. In thisregard, they should adopt a well-defined methodology during the ontology development process inordertodefine areliableandefficientmodelthat buildsknowledge ofthetargetdomain.Fourwell-known methodologies can be selected from a diverserange ofmethods that were offered for ontology construction, namely methontologymethodology[24],101method[25],Neonmethodology[26]andagilemethodology[27].

• Evaluation Techniques:Thiscriterion referstothesuggestedmethodsandtechniquesusedforevaluatingtheseworks.

Itissuggestedtoaddressthisquestion”Howdotheauthorsevaluatetheirmodels?”.

Infact, itisaprimordialtasktoevaluatethequality,performanceandusefulnessoftheproposedontology.Therefore, fourwell-knowntechniquescanbeusedforontologyassessment,namelygoldstandardassessment,humanevaluation, data-driven evaluationandapplication-basedevaluation[28].Goldstandard evaluationaims tocompare theproposed ontologytoeitherahigh-levelmodelordefinedstandardsintheconcerneddomain.Humanevaluationmethodisbased on severalpredefined comparisoncriteria that are proposed to evaluatethe ontology design, such asits clarity level, completeness,consistency,Gruber[29]and Gómez-Pérez[30].Application-based evaluation consistsin using an ontology inaspecificapplicationinordertoevaluateitsresults.Thisontologycanbecomparedtoapredefineddatasource,such asacorpusofdocumentsinaparticulardomain,calleddata-drivenevaluation.

4. Semantic-basedapproachesforIoT

Several projects are focusing on addressing the semantics IoT like IoT-A [31], SOFIA⁷, SemSorGrid4Env¹, Linksmart⁸, IoT.est⁹,openIoT³,FED4FIRE¹⁰,Vitalontology¹¹andCityPulse².

Gyrardet al [32], havedeveloped a linked open vocabulary project(LOV4IoT¹²) that collects andregroups numerous relevantontologiesfortheIoT domainsuch as,SAREF¹³standardizedbyETSI withthenameSmartM2M,spitfire¹⁴andthe OneM2Mbaseontology¹⁵.

Inthefollowing sub-sections,we proposean overviewofthe realizedresearchesin thistopic(the37 selected papers asitwasclarifiedintheprevious section)thatwere classifiedaccordingtotheir maincontributions.Wealsocomparethe proposedworksineachsub-sectionbasedonourresearchquestions.However,Tables3and4highlightsthe mainfeatures anddrawbacksofthesestudies.

4.1. IoTDatarepresentation

ThestudiesinthiscategoryfocusontheuseofSWTinrepresentingthesemanticsofIoTresourcesandtheirdata. In2012,theSemanticSensorNetworkIncubatorGroup,belongingtotheWorldWideWebConsortium(W3C),developed anontologycalledSemanticSensorNetwork(SSN)[9].Thisontology isdevelopedthroughthereviewofdifferentproposed ontologies,namelySemSOSontology[67],Ontonym-Sensor[68]and CESNontology[69].The SSNontology describes sensors intermsofcapabilities,measurement processes,observations anddeploymentsinordertodefine thesemanticinteroper- abilityofphysicalsensornetworks. Itscoreconcepts are sensorsandtheir featuresandproperties,observations,systems, measuringcapabilities,operatingandsurvivalrestrictions,anddeployments.

7http://sofia2.com/

8https://linksmart.eu/redmine/projects

9http://ics-iot.weebly.com/iotest.html

10 http://www.fed4fire.eu/

11 http://vital-iot.eu/

12 https://lov4iot.appspot.com/

13 https://sites.google.com/site/smartappliancesproject/ontologies

14 http://sensormeasurement.appspot.com/ont/sensor/spitfire.owl

15 http://www.onem2m.org/ontology/Base

Table 3

Comparison between Semantic-IoT related works.

Items Assets Drawbacks

[9] - This work, proposed an ontology for sensor networks and their observations (SSN ontology).

- It was based on other ontologies, like O&M and SemSoS.

- It is available online.

It was used in diverse projects, such as IoT.est, open IoT,FED4FIRE, etc.

- The work was proposed for only sensors

- It didn’t consider the modeling phase of sensors services.

- Evaluating the semantic quality (consistency, clarity, coverage, etc.) of the SSN ontology was not addressed.

- Privacy and security issues were not presented and interpreted.

- It did not support reasoning mechanisms for sensor configuration and data management.

[33,34] The SSN ontology was extended to wireless and cloud sensor networks respectively.

- They focused only on sensors - They are not available online.

- They did not define sensors and data management techniques.

- Lack of privacy and security issues.

[35] - This work expanded the SSN ontology to cover actuator devices and came up with SOSA ontology.

- It is available online

- Both sensor and actuator services were not modeled.

- It did not support reasoning mechanisms - Semantic quality was not assessed.

- Privacy and security issues were not addressed.

[36–40] These works have modeled IoT devices (sensors, actuators, RFIDs) and their data

- They were based on the SSN ontology, except [39] and [40]

which were built from scratch.

- They were tested based on specific applications (smart city, smart home, healthcare).

- They are available online, except [40]

They were interested only in IoT devices and their data stream without focusing on their services.

- They did not support reasoning.

- The evaluation step did not emphasize semantic quality.

- Privacy and security aspects were not considered.

[11,41,42]

- These works introduced ontologies that describe connected devices (sensors, actuators) and their services.

- They were linked to other SSN and OWL-S ontologies, except [42] . They were tested based on real objects.

- IoT data was not considered in these works.

- They did not concentrate on IoT service selection.

- They did not support reasoning mechanisms.

- The ontology proposed in [11] is the only available online.

[43] -This work proposed a modular ontology for IoT and their services

- Despite that, it is one of the first attempts in IoT, the proposed ontology cover concept about the sensor, actuator, physical object and so on.

- Defined mathematical rules modeled in the ontology for IoT services selection

-The evaluation was missing -The ontology not available online

- Rules for IoT data and resource management were not defined

[44–46] - These works put forward a semantic representation of IoT resources and their services.

They proposed an IoT service selection method based on QoS.

- The developed ontologies were built by reusing SSN and OWL-S ontologies, except [46] which was built from scratch. The suggested ontologies were tested based on a real use case (weather monitoring).

- These ontologies did not consider IoT resource data.

Diverse characteristics of IoT resources were not considered in these works such as connectivity, virtualization, mobility, energy, and life cycle.

- They are not accessible online.

- Rules for inference and reasoning purposes were not presented.

- The semantic quality of the suggested ontologies was not evaluated.

- Security aspects were not taken into account.

ItisthenusedbyseveralprojectslikeSemsorGrid4Env¹⁶,CityPulse¹⁷andopenIoT¹⁸.Thisontology isthenextendedby Bendabboucheetal.[33],forwirelesssensor network(wirelesssensornetworkontology),andbyMullerforsensorcloud (sensorcloudontology)[34].

Theabove-mentionedworkshavemainlyfocusedonthesemanticrepresentationofsensorsandtheirobservations.

TheMELODYprojectsintroducedasemanticactuatornetwork(SAN)¹⁹torepresentthesemanticsofactuatorsandtheir capabilitiesandroles.Besides, theSOSA ontology [35],proposed by the jointgroup WorldWide Web Consortium(W3C) andtheOpenGeospatialConsortium(OGC),representstheinteractionbetweensensors,observations,actuatorsandsample concepts.

Recently, theSSN ontology hasbeensimplified byremoving some concepts includingstimulus, systems,measurement andsystemcapabilities,andbyextendingtheSOSAontologytorepresentknowledgeaboutactuators[35].

Nonetheless,theIoTisnotonlycomposedbysensorsandactuators.Fromthisregard,Gyrardetal.[36]proposedanM3 ontologythatisextendedfromtheSSNontologythatincludesvariousconcepts,suchastransducer,RFIDStag,andcontroller. ThemeasurementconceptwasproposedinthismodelinordertorepresentdatastreamsobtainedfromIoTresources.This

16http://mayor2.dia.fi.upm.es/oeg-upm/index.php/en/activeprojects/

17http://www.ict-citypulse.eu/page/

18http://openiot.eu/

19https://www.irit.fr/recherches/MELODI/ontologies/SAN.html

Table 4

Comparison between Semantic-IoT related works (continued).

Items Assets Drawbacks

[47–49] - These works proposed a context-awareness IoT service descriptions and selection approaches.

- They reused existing ontologies, such as SSN and OWL-S.

- They suggested context-awareness IoT service selection and discovery (temporal selection, geographic selection, uncertain selection, etc.).

- They did not present IoT data in their ontologies.

- Management of IoT devices and their services through reasoning rules was not defined.

- The evaluation was either missing or poorly described.

- The proposed ontologies were not accessible online.

- Security issues were not considered.

[50–54] - These works proposed a distributed-based solution for describing and selecting IoT resources and services based on SWT.

- The suggested method was evaluated based on specific applications.

- Semantic modeling and utilisation of IoT resource information were not addressed.

- The evaluation focused only on the application-based method without addressing the criteria-based method to check the coverage, clarity, consistency, and completeness of these ontologies.

- Reasoning and interpreting IoT resources and their services were not defined.

- The proposed ontologies are not available online.

- Security domain knowledge was not presented in these ontologies.

[10,55–

58]

- These works proposed monolothic ontologies for the semantic representation of IoT resources, information and services.

- Most of these works followed the application-based method as an evaluation tool.

- Most of works were linked to other ontologies.

They addressed only modelling and annotation issues and ignored the interpretation and reasoning phases.

- Only ontologies in [57] , [58] are accessible online.

- Evaluating the semantic quality of the suggested ontologies was missing in all approaches.

- Security domain knowledge was not presented in these ontologies.

[59–62] - These works set forth security ontologies for IoT applications.

- The proposed ontologies were linked to the previous ones that were specific to sensors, networks, web, etc.

- Only [59] and [60] are available online.

- Reasoning mechanism to manage the occurred attacks was only defined by [61] and [62] .

- Evaluation concentrated on implementing ontologies in applications without taking into account the quality of these ontologies, such as their correctness, completeness, clarity, and consistency levels.

- The reuse of IoT ontologies, such as IoT-O, IoT-lite, and so on was treated only by [61] .

[63] , [64–66]

- These works introduced new concepts to define WoT semantics, such as WoT resources, Identifier (IRE), application protocols (COAP, HTTP), etc.

- The majority of these works evaluated their ontologies within a web application.

- These ontologies are available online.

- Evaluating the quality of the developed ontologies was not highlighted.

- Reusing IoT domain knowledge (IoT devices,data, services) in these ontologies was absent.

- Security aspects were not considered in these ontologies.

ontologywasthensimplifiedinthecontextofFIESTA-IOTH2020EUprojectanditwascalledM3lite²⁰.Itwasintegrated inaframeworktofacilitatethedevelopmentofIoTapplications.

Todealwiththereal-timesensordata,theauthorsin[37]suggestedaframeworkforreal-timesemanticannotationof streamingIoTdatatosupportdynamicintegrationintotheWeb.Thisframework(calledCityPulseframework)wasproposed in the context of the CityPulse² project.It is based on a Stream AnnotationOntology (SAO).The main concepts of this ontologyarestreamdata,streamanalysisandstreamevent.ThisontologywasextendedbyElsalahetal.in[38]byadding otherconceptstorepresentIoTdevices,time,location,dataunits,andvalues.TheextendedversionwascalledtheIoTdata streamsontology.

ToensureinteroperabilitybetweentwoormoreIoTdatahubs,Tachmazidisetal.[39],presentedasemanticenrichment oftheBT HypercatData Hub.The lattergathersdata fromdifferentsources tobe broughtonto a commonplatformand presentedtousersanddevelopersinaconsistentway.Therefore,themainconceptsoftheBTHypercatontologyaresensor stream,eventstream,sensorfeed,eventfeed,etc.TheaccesstotheHBHypercatontology isensured throughtheSPARQL language,basedonthemappingbetweenSPARQLandSQLqueries.

Inaddition,theauthorsin[40]haveproposedanontology-drivenapproachtoenableautomaticgenerationoffirmware forthe IoTdevicesandmiddlewarefortheapplicationsthroughhuman-machine interfaces.Thisontologyiscomposed of sixmodules.ThefirstoneiscalledtheprogramminglanguagesI/Ostructureontology isusedtomakethesourcecodesof thedevelopedfirmwareandapplicationsunderstandablebytheusersthroughtheinput/outputdatastructuresdescription.

The data types ontology represents all data types supported by the system. The visual objects ontology is proposed for describing and monitoring the available visual objects and visualizing the graphical scenes of the resultof IoT devices.

Thesemanticfilters ontologyis interestedinraw datafilteringandpre-processing(e.g.denoisingsensordata,andfusing data).Theelectroniccomponentsandmiddlewareontologydefinesadescriptionofdifferentprogrammabledevices(sensor, actuator,transducer,etc.)andadescriptionofdifferentprogramminginterfacesofthedevelopedsystem.

Table5,showsthattheseapproachesutilizeSWTtoformalizetheirontologies.Inaddition,wenoticethatalmostofthese approachestaketheadvantageofreusingtheexistingontologiesinsteadofbuildingtheironesfromscratch.Therefore,only

20http://lov.okfn.org/dataset/lov/vocabs/m3lite

Table 5

Summary of semantic-based approaches for IoT resources and data description.

Reference RQ2 RQ3 RQ4 RQ5 RQ6 RQ7

CoT CoL CoI CoTr CoR

[9] OWL2 SemSOS, Ontonym-Sensor, CESN, O&M × × × × application

[33] × SSN × × application

[34] RDF, SPARQL SSN × × × × application

[35] OWL SSN, DOLCE, × × × × application

[36] OWL SSN × × Methondology application

[37] OWL Dul and SSN × × × × × × application

[38] RDF, SPARQL SSN, SOSA, IoT-lite × × × × 101 methodology application

[39] RDF, OWL, SPARQL, JSON-DL × × × × × × × × application

[40] OWL × × × × × × × application

two works[36,38] followa methodologyduringthe developmentprocess.Furthermore,only theworks [9,33,35] present their ontologiesinamodularmodelsinordertofacilitatetheirextension andreuse.The mainmodeledcontexts inthese works are the time and the location with lessattention about the interconnectivty, the requirement and the trajectory contexts. Concerningtheevaluationoftheir ontologiesall thestudied approachesinthiscategoryfollowthe application- basedevaluation.

4.2. IoTServicesrepresentation&discovery

Inthissection,wereviewrelatedworksaboutdescribinganddiscoveringIoTresourcesandtheirservices.Thesuggested ontologiesinthiscategorydefineconceptsonlyforIoTresourcesandtheirservices.

Deetal.[41]presentedasemanticmodelforIoTcomponents(entity,resourceandservice).Theyproposedanextension oftheOWL-SontologyforIoT.Thereby,theyaddedtheconceptofIoTserviceasasubclassoftheOWL-S²¹ ontology.

In thecontext ofthe ADREAMproject,the authorsof[11] proposed a modularontology (IoT-O)²² to ensuresemantic interoperabilitybetweenIoTcomponents.IoT-Oincludesseveralmodules,likesensing,actuating,lifecycle,serviceanden- ergy modules.The sensingmoduleextendsseveralconcepts fromtheSSN ontology.The actuatingmoduledescribeshow thesysteminteractswiththephysicalworldandismodeledwiththeclassesactuator,actuation,etc.Thelifecyclemodule models thestate ofmachinesanddevice usages. Theservicemodule, showingweb serviceinterfaces,consistsofservice, operationandmessageclasses.TheenergymoduleprovidesclassesforrepresentingthepowerconsumptionofIoTdevices.

Theauthorsin[42]suggestedasemanticserviceontology toensureheterogeneousIoTservicedescriptions indifferent contextsorplatforms.Theproposedontologymainlycomprisesaservice-objectconceptthatismadeupofthreesub-classes:

property,capability,andserverprofile.Propertyrepresentsthestaticstatesoftheconnectedobjects.Thecapabilityconcept representsthedynamicallygenerateddatabyobjects.Theserverprofiledeterminesaconfigurationofphysicalobjectswhen itinteractswithspecificplatforms.

AuthorsinthepreviousstudiesdidnotperformanymechanismforIoTservicediscovery.Toovercomethisissue,many proposedapproachesaredividedintofourcategories.

• Mathematical-basedapproachthatisbasedonmathematicalequationstoretrieveIoTservices.

• QoS-basedapproachconcentratesonprovidinghighqualityservices.

• Context-basedapproachaimstosuggestservicesforusersbasedonapredefinedcontext.

• Distributed-basedapproachallowstodiscoverheterogeneousanddistributedIoTservicesinascalableway Thesecategoriesaredetailedinthefollowingsub-sections.

4.2.1. Mathematicalbasedapproach

Oneofthefirstattempts,that proposedsemanticsIoT,issuggestedbyHachemetal.[43],whointroducedanIoT mid- dleware based on the service-oriented paradigmthat abstracts things as services.This middleware includes three major ontologies (device,physicsandmathematicsandestimation). Thedevice ontology aims todescribe ”things”. Thephysical ontology permitstomodelnotonlyrealworldentitiesasphysicalconcepts butalsomathematicalformulasandfunctions tofacilitatetheIoTservicediscovery.Theestimationontologydescribesmodels,whichcanbeusedwhenaserviceisun- available.

4.2.2. QoSBasedapproach

ThissectiondisplaysapproacheswiththemaingoalofdiscoveringandselectingIoTservicesbasedontheQualityofSer- vice(QoS)values.Theauthorsin[44]haveproposedacommonconceptualmodel,named”PhysicalServiceModel(PMS)”,

21 https://www.w3.org/Submission/OWL-S/

22https://www.irit.fr/recherches/MELODI/ontologies/IoT-O/

fordescribingheterogeneousIoTservices.ThePMSiscomposedofthreemainconcepts,namelydevice,resource,andservice. The device classrepresents the employed hardwarethat isattached toa real-world entity.The resource concept defines a computational element, which is hosted on a device and exposed by a servicethrough a common interface. The ser- vice concept represents the services of IoT devices, such as identifying,actuating and sensing. The PMS model contains spatio-temporalfeaturesthatcharacterizethedeploymentofthesethreeconcepts.ThesefeaturesarethenexploitedasQoS attributesforIoTservicerating.Otherattributesaresuggested,suchasreliability,reputation,andexecutioncost.Finally,the serviceselectionofthismodelisbasedontheuser’spreferencesandrequirements.

Theauthorsin[45]havesuggestedasemantic-basedframeworkforIoTtransactionmodelingandprocessing.Thisframe- work is composed of various layers:the entity link layer, the semantic annotation layer, the service registry center, the transactionconstructionlayer,andthetransactionexecutioncontrollayer.ThisframeworkisbasedontheOWL-Sontology asitisthebasisforlogicalreasoning.Thisontologyisextendedbyaddingtheservice-statusconcepttorepresentdynamic servicesforIoTentities.Thisconcepthasnotonlyreferstoalocation(currentlocation)andthecurrentstatusofIoTdevices, butalsoasequenceofIoTtransactions.Inaddition,Yachiretal.[46]proposedasemanticmodelforsmartobjectsandusers requestresolutionintheIoT domainusingontology anddescriptionlogictechniques.Thismodelfacilitates thereasoning overservicesanddevices.TheselectionofservicesisbasedontheQoSlevel.Infact,onlyservices,providinghigherquality thantherequiredQoSlevel,arereturnedtotheuser.

4.2.3. Contextbasedapproach

Inthissection,wedescriberelevantworksthatmainlyfocusedonhowtoprovideIoTservicesaccordingtoapredefined context.Accordingly,asignificantworkhasbeenproposedbyNambietal.in[47],whichsuggestaunifiedsemanticknowl- edgebaseforIoTthatconsistsofseveralontologies(resource,location,contextanddomain,policyandserviceontologies).

Theresourceontology representsanentityinIoT(e.g.,sensor,actuator,physicalobjectandcompositeobject).Thelocation ontology describessemantic geospatialinformation forIoT.Thisontology extendsthe GeoNamesontology²³. Thecontext anddomain ontologies represent contextual information anddomain-specific knowledge. The policy ontology isused to provideinformationonhowtoaccomplishaservicerequestedby anactorindynamicenvironments.Theserviceontology describes,representsandmodelsthe IoTservices.In[48],theauthorshaveprovided anontology basedcontext modelto representuncertainandtemporalcontext.Therefore,DynamicBayesianNetworks(DBN)areadoptedtoreasonaboutthese contextsforuncertainIoTservicesdiscovery.Besides,theworkdonebyLietal.in[49]developedadecentralizedLocation- preservingcontext-awarediscoveryframework,whichisbasedonSWT.Itusesontologytoencodecontextinformationand matchquerieswithservicestoselectthemostappropriateones.

4.2.4. Distributedbasedapproach

Thestudied approachesin thissection focused on applyingthe distributed-basedsolution forIoT services description andselection.Theauthorsin[50] havedevelopedalightweightontologyforIoTservicesdescription,whichcontainsseven mainmodules:IoTservice,servicetest,Qualityofservice(QoS)andqualityofinformation(QoI),deploymentplatformand networking, observation and measurement, IoT resource, entityof interest andphysical locations. Scalable access to IoT servicesand resources isrealized through a distributed,semantic storagedesign. In[51], authorsset forth an integrated semanticserviceplatform.ThemajorpurposeofthisplatformistoresolvethreemainproblemsindistributedIoTdomains byapplyingsemantictechnologiestoIoT,suchassemanticdiscovery,dynamicsemanticrepresentation,andsemanticdata repositoryforIoTresources.ThisplatformisbasedontheIoT-basedserviceintegrationontology.The lattercontainsthree mainconcepts:servicethatrepresentsoneormoreIoT-basedservices,userthatdescribesinformationaboutend-users,and referenceclassindicatesthewaytorefertoIoTresourceslocatedinexternalIoTserviceplatforms.

Theauthorsin[52] proposedForwardDS-IoTafederated discovery serviceintheIoT context,whichincludesa seman- ticmodel.Thelatteris basedonexisting ontologies,such asSSN,SAN,andOWL-Sontologies.Additionally,Willneretal.

[53]haveputanopen-multinetupperontology(MON) tosupporttheinteroperabilityoffederated infrastructures.Itfacil- itatesthe management ofheterogeneousresources through theamalgamation ofdifferentsemantic ontologies.The MON ontology is composed of six sub-ontologies:omn-federation, omn-resource, omn-service, omn-lifecycle, omn-component, omn-policy,andomn-monitoring.Bythesametoken,Zhuetal.[54]havesuggestedanSOA-basedsolutionforIoTservices discovery.In fact,they definedan ontology forcyber-physicalsystemsand IoT domains tofacilitate theserviceselection and composition. This semantic model is extended from the OWL-S ontology by applying the OWL-S constructs of the process-profile-grounding. The PT-SOA ontology includes four main concepts: the physical profile is used to describe the characteristics,components,andconstituentsofphysicalthings;theoperationprofileisproposed tocontrol physicalthings oroperationswhenprovidingaservice(maintenance,constraints,etc.);contextisutilizedtorepresentthedynamicstateof IoTthingsandthescheduledservicesthatdescribetheservicedeliverycontext(time,location,etc)bythesethings.

Tables 6–10 provide a comparative analysis of the studied works for representing anddiscovering IoT services. They showthattheseapproachesuseSWTtoformalizetheirmodels.Onlyafewworks[11,47,50]usemodularrepresentationfor theirmodels.ThedeploymentcontextsofIoTdevicesarepartiallyaddressed.Moreover,onlytheworkachievedin[11]has adoptedamethodologyduringthedevelopmentoftheir models.Theapplication-basedmethodisfrequentlyusedforthe evaluationpurposes.

23http://www.geonames.org/ontology/documentation.html

Table 6

Summary of semantic-based approaches for IoT services representation.

Reference RQ2 RQ3 RQ4 RQ5 RQ6 RQ7

CoT CoL CoI CoTr CoR

[11] OWL SSN, DUL, PowerOnt, OWL-S × × × NeON Methodology application

[41] OWL SSN ontology, OWL-S × × × × × ×

[42] JSON-LD × × × × × × × × application

Table 7

Summary of semantic-based approaches for IoT services discovery: mathematical-based ap- proach.

Reference RQ2 RQ3 RQ4 RQ5 RQ6 RQ7

CoT CoL CoI CoTr CoR

[43] OWL × × × × × × × × ×

Table 8

Summary of semantic-based approaches for IoT services discovery: QoS-based approach.

Reference RQ2 RQ3 RQ4 RQ5 RQ6 RQ7

CoT CoL CoI CoTr CoR

[44] OWL, RDF, SPARQL SSN, Geonames, FOAF, OWL-S × × × × × application

[45] OWL OWL-S × × × × × application

[46] Description Logic × × × × × × × application

Table 9

Summary of Semantic-based Approaches for IoT services discovery: context-based approach.

Reference RQ2 RQ3 RQ4 RQ5 RQ6 RQ7

CoT CoL CoI CoTr CoR

[47] OWL SSN, OWL-S × × × × × ×

[48] OWL × × × × × × × ×

[49] OWL contexts ontologies × × × × × application

Table 10

Summary of semantic-based approaches for IoT services discovery: distributed-based approach .

Reference RQ2 RQ3 RQ4 RQ5 RQ6 RQ7

CoT CoL CoI CoTr CoR

[50] OWL OWL-S, SSN, Geonames, etc. × × × × ×

[51] Jena, OWL, RDF, Pellet, SWRL, SPARQL × × × × × × application

[52] OWL, SPARQL OWL-S, SSN, SAN, Geonames × × × × × × application

[53] OWL, SPARQL, × × × × × application

[54] OWL OWL-S × × × × × application

4.3. IoTData&servicesrepresentation

Thissectionstatestherelevantcontributions,whichfocusonapplyingSWTtorepresentthesemanticsofheterogeneous IoTresources,theirdataandtheirservices.

The authors in[55] suggested an IoT domain modelthat is composed ofdiverseconcepts, such as augmentedentity, user,device,resource,andservice.

Kotis et al.[10], described an ontology that represents knowledge about the IoT devices in order to cooperate with eachotherinalarge-scale,afederatedandcoordinatedway.Themainconceptsofthisontologyaredevice(sensingdevice, actuatingdevice,attacheddevice,computingdevice,etc.),IoTentity,andsomeconceptsareextendedfromtheSSNontology, suchassensor,featureofinterest,observationandproperty.

Ma etal. [56] set forth thea framework fora semantic informationmodel forIoT applications. Infact, the proposed ontologydescribes(i)realwordentities,namelytheobjectbeingmonitored,sensordevicesandthenetworkinfrastructure, (ii)thespatialandtemporaldimension,(iii)thecaptured(dynamicandstatic)data,(iv)servicesincludingapplications(e.g., intheareasofhealthcareortraffic),functionsandinterfaces.