Abstract
Smart communities have recently gained much attention. Researchers have been trying to tackle a number of challenges faced by smart communities. Interoperability is one key challenge that occurs due to different systems using different knowledge representations. To solve interoperability problems, ontologies are seen as a promising solution as they provide a commonly agreed vocabulary for representing data that are understandable by stakeholders of smart communities. Smart communities make use of Internet of Things (IoT) and ubiquitous networks to support communication among objects and devices in such environments. Smart campuses are examples of smart communities. Recently, many articles related to ontologies focusing on smart communities and smart campuses in IoT environments, have been published. This paper presents a Systematic Literature Review that has been conducted using Google Scholar. 18 ontologies for smart communities/smart campuses have been identified and analyzed out of 341 articles from year 2010 to 2019. The review classifies the ontologies in terms of domain, ontologies being reused, availability online, limitations, language adopted and coverage. It additionally discusses on the standards, the level of expressiveness, the ontology development approaches and methodologies adopted by the identified ontologies. Our analysis shows that the identified ontologies have been developed based on different ontological commitments. None of them have come up with a core semantic model that models different collaborating domains in a smart campus such as smart learning, smart management, smart governance, smart room, smart health, smart library and smart parking among others and that enhances cross-domain interoperability in a such an environment. Further details on our findings are presented and discussed in the paper.
Introduction
IoT-based smart campuses are smart communities, which consist of group of interconnected objects that interact with each other over ubiquitous networks (Kinkar et al., 2016). The IoT paradigm grew from ubiquitous computing and now evolves into a model consisting of a number of loosely coupled objects (Korzun et al., 2015). The objects will be able to see, hear, think and perform various jobs (Al-Fuqaha et al., 2015). They are called “smart objects” as they have the ability to sense the environment and use this context information to make appropriate decisions (Gubbi et al., 2013; Korzun et al., 2015). They use ambient intelligence to understand their environments and people and reasoning techniques to make appropriate decisions through the IoT (Al-Fuqaha et al., 2015). Ambient intelligence is defined as a discipline that incorporates artificial intelligence to make the environment responsive and sensitive to actions of people and objects and cater for their needs (Aarts and Wichert, 2009; Cook et al., 2009). Smart objects aim to deliver smart services to their stakeholders consisting of humans, physical objects, cyber world and social world among others (Whitmore et al., 2015). Smart applications are driven by such objects that help in data collection and representation and support real time decision making. Technologies such as Internet of Things (IoT), cloud computing, wireless networks, mobile terminals, RFID and NFC are used to capture and store data in an IoT-enabled smart campus (Nie, 2013; Muhamad et al., 2017). Presence of numerous applications in an IoT-enabled campus imposes the utilization of heterogeneous IoT objects, leading to interoperability problems (Abuarqoub et al., 2017; Korzun et al., 2015). Heterogeneous devices generate a huge amount of raw sensor data and these data are in different formats, types and semantics (Elkhodr et al., 2016). They may need to exchange data without knowing how they are represented (Jara et al., 2014). Given that the smart campus community is highly dynamic, the collected data change from time to time (Adamkó et al., 2014). There is therefore a need to represent information properly in a smart campus. Furthermore, a standardized model for description of data generated by IoT devices is required (Abuarqoub et al., 2017). Fleiner et al. (2017) emphasize the importance of establishing a common data model “that enables the interconnection of fragmented data from heterogeneous data sources”. It is desirable that data and services in a smart campus be expressed in formalized ways and represented semantically in order to be interoperable.
There are a number of types of models and data structures that can be used for semantic data modeling namely key-value, markup scheme, graphical models such as UML, object oriented models, logic based and ontology based models (Strang and Linnhoff-Popien, 2004). Among these models, the ontology based model provides a good formalization language with logic inference ability and is best suited to solve interoperability problems (Ma et al., 2014). The use of an ontology will thus enable integration, communication and coordination of service providers and their clients (Jasper and Uschold, 1999; Korzun et al., 2015). Guillemin et al. (2014) emphasize the need to address semantic interoperability of IoT applications and to use proper guidelines for interoperability enforcement. The European Research Cluster on the Internet of Things (IERC) AC4 released in March 2015 provide a set of best practices and recommendations for semantic interoperability (Gyrard et al., 2015). They emphasize the need to overcome the following challenges:(1) a unified model for IoT data annotation, (2) reasoning mechanisms, (3) linked data approach, (4) horizontal integration with existing applications, (5) design lightweight versions for constrained environments, and (6) alignment between different vocabularies.
To investigate existing ontologies being used in a smart community such as a smart campus, the Systematic Literature Review (SLR) methodology has been chosen. The rest of the paper is structured as follows: Section 2 describes the core concepts of the paper, namely smart campus and ontologies as a semantic model. Section 3 describes the steps of the SLR based on Kitchenham and Charters (2007) guidelines to plan, conduct, and report the review. To plan the review, relevant research questions were identified based on Petticrew and Roberts (2008) guidelines and Google Scholar was chosen as the data source. To conduct the review, a qualitative assessment was carried out. The review reported details of eighteen ontologies. The ontologies are classified in terms of domain, availability online, ontologies being reused and limitations in Section 4. The coverage of the ontologies is classified in terms of common concepts used by the identified ontologies in Section 5. Section 6 highlights the ontology language and the standards adopted by the identified ontologies. Section 7 discusses on concepts related to level of expressiveness of the ontologies. The approaches and methodologies for ontology development adopted by the ontologies are presented in Section 8. Limitations of the review are described in Section 9. Finally, Section 10 concludes the paper.
Concepts
This section introduces the background literature by describing the main concepts related to the paper namely smart campus and ontology.
Smart campus
A smart campus is a smart community that is also known as an intelligent campus. A modern campus is “a collection of buildings and grounds that belong to a given institution, either academic or non-academic” (Campus, 2020). Googleplex and Apple Campus are examples of modern campuses (Campus, 2020). Though the term ‘campus’ may belong to non-academic institutions, it is most often used to refer to a university environment which is associated with teaching and learning. Tian et al. (2018) refer to a smart campus as one that “uses technology and infrastructure to support and improve its processes in campus services, teaching, learning, and research”. Yang et al. (2018) define a smart campus as “an intellectualized campus work, learning and living integrated environment, which is based on IoT”. Prandi et al. (2019) define a campus as a term used to “identify buildings and ground, or more generally places, where a university is situated”. Fortes et al. (2019) define campuses as “small cities, where achievable improvements to its management, governance, sustainability, and learning activities are enormous in terms of energy and water efficiency, emissions, mobility, health and well-being, nature, and teaching”. Villegas-Ch et al. (2019) refer to a smart campus as one that “allows a better coexistence between the university population and its surroundings, adequately manages the resources within the campus, and provides favorable places for learning”. Dong et al. (2020) refer to a smart campus as a smart education system with innovative education concepts. In recent years, smart campuses have gained much popularity and there is a growing interest to carry out more research in that field. Researchers have come with innovative solutions to tackle different problems in smart campuses. Valks et al. (2020) have identified the process-level requirements to make strategic decisions in smart campus management. Feng (2020) has proposed a smart campus system model based on big data analysis that aims to improve the efficiency of information operation in a campus. Moraes et al. (2020) have proposed a framework that will collect and process sustainability data in a smart university campus, along with a simulator for a green smart campus system. An and Xi (2020) adopt sustainable design to construct a service system framework for a smart campus.
A smart campus, which normally consists of buildings, libraries, classrooms, recreation areas and student centers amongst others and stakeholders such as students, teachers and management teams, incorporates elements of campus intelligence (Kwok, 2015; Nagamalla et al., 2018). Ng et al. (2010) identify six domains for a smart campus namely iLearning which supports students and faculty acquiring knowledge, iManagement for physical aspects of a campus such as smart building management, iGovernance which deals with organizational aspects of the campus, iSocial which handles the social networking and communities of a campus, iHealth which tackles health issues such as epidemic alert systems and iGreen which is concerned with aspects of green ICT and sustainability and smart energy harvesting. In terms of infrastructure, a smart campus can be described by several parameters such as smart education, smart room, smart parking (Sari et al., 2017). There are a number of systems available in a smart campus namely learning management systems, student record systems and facility management systems amongst others. These systems have to be linked and integrated in order to provide useful information and thus increase the level of smartness in a campus environment (Kwok, 2015). To facilitate system integration, data from the various systems have to be interoperable. Hirsch and Ng (2011) additionally highlight that different services in a smart campus have to be integrated as well so that they can communicate with each other. To allow services to interact, it is important that they “understand” the data they have to consume and process (Hirsch and Ng, 2011). To ensure interoperability, a standard-based representation of services should be considered (Conde et al., 2012; Karavirta et al., 2013).
Ontology as a semantic model
There are several definitions for ontologies in literature. Gruber (1993) defines an ontology as an “explicit specification of a conceptualization”. Borst et al. (1997) state that an ontology should be formal, agreeing, with the general notion above and should additionally include the aspect of sharing. Studer et al. (1998) extend this by stating that the ontology must be explicit and thus defined an ontology as a “formal, explicit, specification of a shared conceptualization”. Uschold (1998) defines an ontology as one that “may take a variety of forms, but necessarily it will include a vocabulary of terms, and some specification of their meaning. This includes definitions and an indication of how concepts are inter-related which collectively impose a structure on the domain and constrain the possible interpretations of terms.”
Ontologies define a commonly agreed vocabulary for representing data of a particular domain, making the data understandable and exchangeable by both humans and machines (Ye et al., 2011). Knowledge sharing and reuse are thus facilitated by using ontologies. Furthermore, ontologies allow the creation of models, which incorporate reasoning and inference capabilities (Abdulrazak et al., 2010). In the IoT domain, heterogeneity of the information coming underlying devices is one major challenge (Agarwal et al., 2016). To allow services using these heterogeneous sources of information to interoperate, Elsaleh et al. (2020) recommend the use of common semantic models that use the same concepts and relationship between concepts. Ontologies are capable of resolving semantic heterogeneity due to the shared comprehension and common language they provide (Elsaleh et al., 2020; Ghawi and Cullot, 2007). Additionally, in case different interpretations and representations are explicitly captured in an ontology, the ontology disambiguates concepts and support interoperability (Zhu and Madnick, 2006). Ontologies are thus used to achieve efficient semantic interoperability among software systems (Ghawi and Cullot, 2007; Jasper and Uschold, 1999). Heflin and Hendler (2000) have defined semantic interoperability as “integrating resources that were developed using different vocabularies and different perspectives on the data. To achieve semantic interoperability, systems must be able to exchange data in such a way that the precise meaning of the data is readily accessible and the data itself can be translated by any system into a form that it understands.” Barnaghi et al. (2012) propose solutions like ontology mapping and matching to link resources from different ontological models and to enhance semantic interoperability. Therefore, the use of ontologies is seen as a promising way to solve semantic interoperability problems.
Methodology
Grant and Booth (2009) describe fourteen types of reviews along with their strengths and weaknesses. Among these fourteen types of reviews, the systematic review of literature (SLR) has been chosen. A systematic review aims for exhaustive and comprehensive searching, adhering to relevant guidelines (Grant and Booth, 2009). Additionally, this approach includes meta-analysis and reduces large amounts of data into a small amount of consistent and precise data that can be used for decision-making (Mulrow, 1994).
This section explains the processes of the SLR, which are based on the guidelines proposed by Kitchenham and Charters (2007). Research goals and objectives of the SLR are specified and the steps of the research methods are explained in detail.
Research goals and objectives
The main goals of the SLR are to:
Firstly, identify and analyze ontologies for a smart campus in the context of IoT and investigate whether there exists a core model that enhances interoperability across different collaborating domains in a smart campus Secondly, describe the coverage of the identified ontologies Thirdly, verify whether the identified ontologies have used standard languages for development or are based on any standards Fourthly, investigate if these ontologies have catered for lightweight versions that fit constrained environments Fifthly, identify the ontology development approach and methodologies used by the relevant ontologies
Research methods
According to the guidelines proposed by Kitchenham and Charters (2007), the systematic review is carried out in three phases: (1) Planning the review (2) Conducting the review, and (3) Reporting the review.
Planning the review
In this phase, research questions are formulated, data sources are identified and relevant criteria for selecting articles in the data sources are defined. To design the SLR questions, criteria by Petticrew and Roberts (2008) are followed as shown in Table 1.
Criteria and scope for research questions
Criteria and scope for research questions
The following research questions have been identified and are intended to be answered during the review:
The motivation behind this question is to study existing ontologies and semantic models belonging to a smart academic campus in an IoT environment. It additionally aims to investigate the objectives and concepts of the ontologies and the domains for which they were designed. As stated in Section 2.1, Ng et al. (2010) have identified six domains of a smart campus, which include iLearning, iSocial, iManagement, iHealth, iGovernance and iGreen. This SLR aims to find out whether there exists a core semantic model that caters for these six domains as well as other important domains in a smart campus and that enhances cross-domain interoperability in a smart campus.
Ontology reuse is defined as “the process in which available (ontological) knowledge is used as input to generate new ontologies” (Bontas et al., 2005). This research question also aims to find out whether the identified ontologies have been designed from scratch or have used any existing ontology. In case ontologies have been reused, it is important to know which ones and the purpose they serve.
Last but not least, it is important to identify the limitations of the ontologies as they may open up new research avenues that may be investigated in the future by experts in the field.
To achieve semantic interoperability in an IoT-enabled smart campus, ontologies need to communicate with each other. Techniques such as ontology mapping and alignment can be used to solve ontology heterogeneities and mismatches (Klein et al., 2002). Ontology mapping and alignment require in depth knowledge about the conceptualizations behind ontologies to be mapped or aligned and their semantic similarities. Therefore, once the ontologies/semantic models have been identified in RQ1, it is of utmost importance to know the coverage of these ontologies, that is, what the common terms and terminologies adopted by these ontologies are so that they can be used. This research question therefore aims to identify these concepts and to classify the coverage of the ontologies based on these concepts.
Ontologies should be implemented using standard ontology languages in order to ensure that the conceptualization is formally and explicitly encoded (Ye et al., 2007). Ontologies represented with semantic technologies, namely Web Ontology Language (OWL), Resource Description Framework (RDF), RDF Schema are used to describe the different capabilities and properties of smart objects along with semantic representation of linked data (Jara et al., 2014; Kiljander et al., 2012). It is interesting to know which language has been mostly used for development for the identified ontologies.
To solve heterogeneity problems between different ontologies and models in the IoT context, a standard ontology that shares and agrees on a common vocabulary and knowledge model for describing data is desirable (Barnaghi et al., 2012; Khriyenko et al., 2013; Serrano et al., 2015). The use of standards helps to achieve interoperability. Giri et al. (2017) highlight that without standards, services provided in an IoT domain will not be accessible to others. Trappey et al. (2017) describe emerging standards for IoT: International Organization for Standardization (ISO), the International Electrotechnical Commission (IEC), the Guobiao standards (GB) and World Intellectual Property Organization (WIPO). The IEC 61360, the IEC 61987, the IEC 62683 (Common Data Dictionary) are three standards that include the related terminology of IoT (Trappey et al., 2017). Additionally, IEEE P2413 is a developing standard for an architectural framework for the IoT. ISO 37101, ISO 37102, ISO 37120 and ISO 37123 establish a set of standardized indicators in the context of sustainable development and resilience of smart communities. The ELOT 1457 standard (Greek) and BS 8904 are other examples of standards that focus on smart communities.
This research question also aims to verify whether the identified ontologies in RQ1 represent or are based on any emerging standards for IoT and smart campuses.
Ontologies vary from lightweight, rather informal, to heavyweight, and formal ontologies (Uschold and Gruninger, 2004; Zhu and Madnick, 2006; Ye et al., 2007). Large and expressive ontologies normally utilize considerable resources for reviewing and understanding the specification (Hetmank, 2014). These ontologies are termed as heavyweight ontologies. They make use of axioms and constraints for interpretation of concepts and relationships (Ye et al., 2007; Zhu and Madnick, 2006). They are considered unsuitable for constrained environments (Bermudez-Edo et al., 2016). Compared to heavyweight ontologies, lightweight ontologies are taxonomies consisting of concepts such as terms and atomic types and hierarchical relationships among the concepts (Ye et al., 2007; Zhu and Madnick, 2006). Lightweight ontologies are less expressive compared to heavyweight semantics (Ye et al., 2007).
Lightweight semantics are recommended for the context of an IoT-enabled smart campus. Researchers argue that in IoT environments, which encompass resource-constrained sensors and devices, heavyweight semantics would not be suitable as they use lots of computation and memory. Such resources may be unavailable in IoT sensors and devices (Barnaghi et al., 2012; Poslad et al., 2015; Serrano et al., 2015). Furthermore, Elsaleh et al. (2020) argue that with the growing number of sensors and data, annotation and query times have become a bottleneck in the real-time processing of data coming from IoT environments. They thus recommend lightweight models with a minimum number of concepts and relationships between the concepts to reduce the processing time of IoT data.
The motivation behind research question RQ4 is therefore to investigate whether the identified ontologies in RQ1 have addressed concepts such as level of expressiveness when being designed.
There are two different approaches that are normally used for ontology development namely top-down and bottom-up. In a top-down approach, the process starts with modeling the concepts and relationships of a particular domain from a very generic level and then proceed to subsequent specialization of the concepts (Noy and McGuinness, 2001; Sure et al., 2002). On the other hand, in a bottom-up approach the process starts with the definition of the more specific classes/objects and their interactions and then move to more general concepts (Noy and McGuinness, 2001; Van Der Vet and Mars, 1998). The hybrid or middle-out approach combines both top-down and bottom-up approaches. Several methodologies for ontology development exist such as TOVE (Gruninger and Fox, 1994), Uschold and King methodology (Uschold and King, 1995), Bernaras methodology (Bernaras et al., 1996), Methontology (Fernández-López et al., 1997), Ontology Development 101 (Noy and McGuinness, 2001), DILIGENT (Tempich et al., 2004), iCAPTURer methodology (Good et al., 2006), GM methodology (Castro et al., 2006) and NeOn (Suárez-Figueroa et al., 2012) among others.
The motivation behind this research question RQ5 is to investigate the approach for ontology development and methodologies adopted by the identified ontologies in RQ1. It would be interesting to find out which approach or methodology is most popular and works best. In the future, we plan to use one of the approaches and methodologies to, eventually, develop a semantic model for a smart campus.
The main data source selected for the search is Google Scholar which includes papers from other data sources such as IEEE, Springer Link, ResearchGate, Semantic scholar, ACM, Wiley Online Library, Science Direct, Digital Library, Victoria Research Archive, Sage Journals, AUT and UCC Library amongst others. The retrieval of documents for the review was based both on manual and electronic search.
The search string consists of the following keywords namely ‘semantic model’, ‘ontology’, ‘smart campus’, ‘smart community’ and ‘Internet of Things’. Connection words AND and OR were used to combine the keywords. The final search string is defined as follows:
The inclusion criteria consist of the following:
Studies carried out in English
Studies relevant to the above keywords from the title and abstract
Peer reviewed journal papers, conference papers and book chapters
Masters thesis and PhD thesis
Articles discussing concepts of ontologies in a smart campus environment
The exclusion criteria consist of the following:
Studies not in English
Duplicate studies reporting same results
Studies where the above keywords appeared only in references
Non peer-reviewed articles
Editorial, abstract or short paper (less than 4 pages)
This section describes how relevant publications were collected, selected and monitored. It shows how data have been extracted from the libraries and synthesized for analysis in the next sections.
The selection process was conducted in three stages:
Selection of publications in the area: Applying the search expression on Google Scholar has provided us with 372 articles. The search was then limited to papers of the past ten years, that is from year 2010–2019 as according to Google Trends (as shown in Fig. 1), it can be observed that as from year 2010, there was a growing interest worldwide in “Internet of Things” and “IoT”. The search has resulted in 341 articles. The results were sorted in descending order of relevance as determined by Google relevance.
Google trends of the search terms “Internet of Things” and “IoT” from 2010–2019.
Preliminary cataloguing of publications: A preliminary filtering of the materials was conducted based on inclusion/exclusion criteria. Citations, duplicates and studies not in English were removed and this resulted in 328 articles. After reading the titles and abstracts of the 328 articles, 53 articles were used at this stage as primary studies relevant to the research. 2 articles were not available in full text, resulting in 51 articles. A database in excel was populated containing the paper name, year, and defined criteria to classify the papers in the database for further analysis. The papers were codified to avoid overlapping.
Qualitative analysis of publications: The 51 articles were analyzed in detail and a qualitative analysis was carried out based on the questions Q1 and Q2 as described in Table 2. Positive answers were valued as one (Y), partly correct answers as 0.5 (P) and negative answers as zero (N). All included articles have scored 1.5 or higher. This resulted in 18 papers, which were included for the study.
Quality assessment questions
The different steps are summarized in the flowchart in Fig. 2.

Flowchart of study selection process.

Number of articles distributed over last 10 years.
Figure 3 shows the breakdown of the 18 relevant articles by the year in which they were published. From the chart, it can be deduced that the number of relevant papers are mostly after the year 2011 as compared to previous years. In year 2016, there are more papers in the field indicating a high research interest worldwide compared to other years.
The findings of the SLR with respect to the five research questions RQ1, RQ2, RQ3, RQ4 and RQ5 are detailed in Sections 4, 5, 6, 7 and 8 respectively.
Ontology characterization
To report on RQ1, this section describes the 18 ontologies identified in the previous section and further, presents a characterization of the ontologies in terms of domain, ontologies being reused, availability online and limitations. It additionally includes a discussion on assessment of the existing ontologies.
Ontologies for IoT-enabled smart campus
Standard Ontology
Abdulrazak et al. (2010) present a standard ontology based on international standards such as World Health Organization (2001), Song and Lee (2008) and WIPO (2008). The ontology presents concepts of environment modeling, being and dynamic aspects and introduces new concepts of referentiality in terms of localization and time. The ontology can be used in a self-organizing middleware that incorporates a fuzzy logic organization reasoning engine.
Top-level Ontology
Ye et al. (2011) present a top-level ontology that covers the following dimensions such as location, time, distance, temperature, humidity and person. The ontology consists of a concept model, context model and an activity model. The concept model uses a generic approach to analyze and structure an information space. The context model handles information acquired from the environment and the activity model represents state of affairs relevant to applications. The ontology can be downloaded at
SSN
SSN ontology is the first W3C standard introduced by the Semantic Sensor Networks Incubator Group (SSNXG) (Compton et al., 2012). The ontology models semantic sensor networks and sensor web applications. It defines a vocabulary for sensors, features and properties, observations, and systems. The ontology is aligned to DOLCE-UltraLite3 (DUL). The SSN ontology is available at
IoT.est
Wang et al. (2012) present the IoT.est ontology, a comprehensive description ontology for knowledge representation in the domain of Internet of Things. The ontology emphasizes essential tasks such as IoT resource and service discovery, IoT service testing, composition and adaptation. The ontology has been designed to be lightweight and integrates aspects of quality of services and quality of information. The ontology is available at
IoT-Ontology
Kotis and Katasonov (2012) present the IoT-Ontology that incorporates existing ontologies such as SSN, DOLCE, QUDT, FOAF and NASA SWEET. The ontology aims to represent interconnected, clustered and aligned smart entities in a unified way so that the entities can support semantic registration, coordination and retrieval of data in an IoT environment. The entities’ metadata are aligned to be used for matchmaking between different entities. The ontology is available at
Campus Scheduling Ontology
The Campus Scheduling Ontology proposed by Fan and Stewart (2014) represents, analyzes and visualizes human mobility and human movement patterns with respect to scheduled activities. In addition to this ontology framework, the authors present a semantic reasoning engine that performs inferences about individual movements and supports aggregated dynamics of the campus. The Protégé-OWL API was used to implement the framework and the Pellet reasoner API was used to implement the reasoning engine. Jena Spatial was used to implement geospatial functions.
SmartLifeOntology
Choi and Rhee (2014) present a base ontology for IoT-Based User-Driven Service named SmartLifeOntology. The ontology consists of four layers namely the place ontology, the object ontology, the context ontology and the service ontology. The place ontology describes the service-domain characteristics such as home service domain. The object ontology, on the other hand, describes sensors and actuators. The context ontology describes a certain phenomenon. The service ontology is comprised of the context and object ontology and aims at describing the service condition and behavior.
OntoIoT
OntoIoT is a two-level semantic framework for modeling IoT applications. It consists of a general upper ontology and application domain ontology, which consist of a set of ontologies for general concepts (Ma et al., 2014). The general upper ontology provides the vocabulary for core concepts of IoT, which include Entity, Dimension, Activity and Service. The semantic model is implemented in OWL using Protégé Editor. The ontology uses ontology-based reasoning to process simple logical inference.
OpenIoT
The OpenIoT ontology formulates and delivers services in the context of IoT (Soldatos et al., 2015). It extends the SSN ontology, which describes the physical and processing structure of sensors. The ontology makes use of cloud resources to process data from sensors. It includes IoT/cloud integration concepts such as annotations for units of measurement, raw sensor values and points of interest. The ontology is available at
SmartRoom Ontology
The ontology for SmartRoom application by Korzun et al. (2015), presents information related to services and participants in a smart room. The ontology presents a vocabulary for SmartRoom services, notifications, participants, materials for SmartRoom users and different activities that are supported by SmartRoom application. Services are offered to participants based on their preference and current context.
Ubiquitous Learning Model
Atif et al. (2015) present a Ubiquitous Learning Model, which aims to incorporate learning practices in a pervasive smart environment. The model takes into consideration paradigms such as context-awareness, ubiquitous learning, pervasive environment and adaptive learning. The model adopts learning resources packaged with the IEEE LOM (Learning Object Meta-Data) standard for integration in the learning environment of the smart campus. This model is extended to Pervasive LOM (PLOM).
Smart Community Ontology
Kinkar et al. (2016) present a Smart Community Ontology that is generic, extensible, interoperable and compatible with heterogeneous data sources. The ontology captures important concepts with respect to events, resources, locations and services. The ontology is supplemented with a number of specialized, highly contextual ontologies to reflect particular use cases. OWL ontology classes are generated from JSON data sources and this generated ontology is mapped to the Smart Community Ontology. The ontology is available at
IoT-Lite
IoT-Lite by Bermudez-Edo et al. (2016) aims to use lightweight semantics to describe key IoT concepts enhancing interoperability and discovery of sensory data in heterogeneous environments. IoT-Lite reuses existing ontologies such as SWEET and SSN. The ontology describes a vocabulary for measurements of devices in terms of quantity kinds and units. IoT-Lite is available at
SemIoT
Kolchin et al. (2016) present SemIoT, a shared ontology that is as high level as possible and that reuses existing standardized ontologies namely SSN and Hydra Core. The ontology consists of several modules that describe devices and components, prototypes and implementations, processes and control commands and an API to access and send data. These modules are organized into smaller ontologies namely SemIoT, Proto Ontology, Hydra PubSub ontology and Hydra Filter ontology. The ontology is available at
IoT-O
IoT-O is an ontology that provides a vocabulary for connected devices and their relation in an IoT environment with the aim of making systems aware of the environment (Seydoux et al., 2016). The development of the ontology follows a modular approach. The ontology reuses existing ontologies namely SSN, DUL and SWEET. The ontology is based on the oneM2M standard (
OLOUD
OLOUD stands for Ontology for Linked Open University Data and has been proposed by Fleiner et al. (2017). It has been developed based on the Uschold and King methodology. It aims to model course information by integrating various sources. The ontology models the following domains namely curricula, subjects, courses, semesters, personnel, buildings and events. OWL 2 RL was used as the formal language for OLOUD. The ontology uses inference rules and class restrictions to make new classifications. The OLOUD ontology is available at
3LConOnt
3LConOnt is a three-level context ontology that can be reused, adapted and extended for different smart scenarios (Cabrera et al., 2019). The upper layer provides a basic taxonomy for high-level context classes. Based on existing contributions in context modeling, the middle layer aims at standardizing the ontological resources. The lower layer defines domain-specific ontologies, which define detailed classes highly dependent on the domain. Lightweight expressivity has been opted to develop the ontology so that it can be easily used and adapted in different use cases. The ontology is available at
SOSA
SOSA Ontology stands for Sensor, Observation, Sample, and Actuator Ontology (SOSA). It is a lightweight event-centric ontology built on top of SSN (Janowicz et al., 2019). The ontology provides a flexible framework to model entities, relations, activities related to sensing, sampling and actuation. The ontology defines common classes and properties for which data can easily be exchanged with SSN ontology. SOSA is available at
Table 3 summarizes the identified ontologies in terms of the year being published, the domain, ontology reuse and availability online.
Ontologies in an IoT-enabled smart campus
Ontologies in an IoT-enabled smart campus
Table 4 highlights the limitations of the identified ontologies.
Limitations
It can be observed from Table 3 that a number of ontologies exist for smart campuses. Ontologies like Campus Scheduling Ontology and Ubiquitous Learning Model are designed for ubiquitous and pervasive environments while SSN, IoT-Ontology, IoT-Lite, SemIoT, IoT-O, SOSA and OntoIoT focus mainly on Internet of Things concepts.
Ontologies are classified in terms of the level of generality into generic ontologies, domain ontologies or application ontologies. While generic ontologies describe general concepts, independent of any particular domains, domain ontologies describe concepts for a particular domain (such as mathematics or chemistry) and application ontologies focus on concepts related to specific applications (Ye et al., 2007). Ontologies such as Campus Scheduling Ontology and OLOUD are domain-specific. Campus Scheduling Ontology focuses mainly on intelligent meeting rooms systems while OLOUD models university courses only. Ontologies such IoT.est, Top-Level Ontology, IoT-Ontology, IoT-Lite, SemIoT, IoT-O, SOSA and OntoIoT are generic ontologies, which cater for mainly observations and measurements data of sensors and IoT devices.
Several methodologies and guidelines recommend ontology reuse in order to develop high quality ontologies and cost-effective (Bontas et al., 2005). Many ontologies reuse existing ones, especially well-known standardized ontologies, to enable simpler integration with other systems (Korzun et al., 2015). Among the eighteen identified ontologies, Standard Ontology, SmartLifeOntology and OntoIoT build their models from scratch and do not reuse existing ontologies. It can be observed from Table 3 that FOAF is being used by a number of ontologies namely Ubiquitous Learning Model, SmartRoom Ontology, IoT-Ontology and 3LConOnt to describe a Person profile. One of the main reasons for its adoption is its simplicity (Bermudez-Edo et al., 2016). OLOUD has used both FOAF and vCard ontology to represent Person. Smart Community Ontology has reused two classes from OpenReferral ontology to model concepts of Organization and AssistanceService. While IoT.est has used OWL-S for service modeling, IoT-O has preferred to opt for Minimal Service Model (MSM) as it is lightweight compared to OWL-S. Dublin Core has been used by OLOUD to model resource metadata while 3LConOnt has chosen CONON and SUMO to model the Resource entity. Temporal aspects are modeled by OWL-TIME and W3C Time and Temporal Aggregates Ontology. While 3LConOnt, SOSA and Campus Scheduling Ontology have opted for OWL-TIME, OLOUD has chosen W3C Time and Temporal Aggregates Ontology instead for modeling recurring events as manually modeling and maintaining DateTimeDescription class in OWL-TIME is time consuming and error-prone (Fleiner et al., 2017). QUDT has been used by IoT-O, IoT-Ontology and SOSA to model units and quantity kinds essential for observation metadata. OLOUD has used iLoc for representing location while 3LConOnt has opted for CONON to model location. While SOSA has reused PROV-O model for modeling events, OLOUD has adopted Event ontology by Raimond and Abdallah (2007) for the same purpose. PowerOnt has been used by IoT-O for energy modeling. It can also be observed that 44% of the identified ontologies in IoT domains have reused SSN for sensor and observations modeling. Its modularity is one of the main reasons for its wide adoption. Upper-level ontologies define abstract concepts for modeling general concepts and properties essential for domains like IoT. Two such examples include DUL and SWEET. SWEET has been adopted by IoT-lite and IoT-Ontology while DUL has been used in SSN ontology, as it is more lightweight than other options (Compton et al., 2012).
Berners-Lee (2006) has pointed out that the Web is a global space consisting of connected documents known as Linked data. Linked data implies that data can be connected to other data sources. One of the principles of Linked data is that it has to be made available on the Web so that it can be reused and referenced by others. Among the 18 identified ontologies, a URI/URL is provided for only 12 ontologies. Out of the 12 ontologies, SSN, SOSA, IoT-Lite and IoT-O have been adopted by other ontologies. Garcia-Castro et al. (2020) keep track of the adoption of SSN and SOSA ontologies in the SSN Usage Document. It can be noted from the document that 23 ontologies have (re) used the SSN/SOSA ontologies. Brain-Computer Interfaces (BCI-O) ontology by José and Méndez (2018) have reused SOSA and IoT-O. IoT-Lite has also been (re) used by a number of ontologies. Internet of Things in Business Processes Ontology (IoT-BPO) has re-used and extended concepts from IoT-Lite (Suri et al., 2017). IoT-Priv ontology, lightweight privacy ontology for IoT, by Arruda and Bulcão-Neto (2019) extends IoT-Lite. I2oTegrator, a Service-Oriented IoT Middleware for Intelligent Object Management, is an ontology derived from IoT-Lite (Abijaude et al., 2018). Additionally, Veiga et al. (2018) have developed a mobile service for context representation using IoT-Lite.
As pointed out by Ng et al. (2010) and Sari et al. (2017), a smart campus consists of aspects of smart learning, smart management, smart governance, smart room, smart health, smart library and smart parking among others. After analyzing the eighteen ontologies from the SLR, it can be noted that none of the identified ontologies cover all the core domains of a smart campus namely iLearning, iManagement, iGovernance, iSocial, iHealth and iGreen as identified by Ng et al. (2010). Apart from these domains, the Key Performance Indicator (KPI) domain is an important aspect that has to be modelled. It represents the measured values, according to a method, in order to monitor the performance of a smart campus. This domain is not included in any of the eighteen ontologies. The KPI domain can be based on the Association for the Set of Smart Campus Indicators Advancement of Sustainability in Higher Education- Sustainability Tracking, Assessment & Rating System (AASHE-STARS) (
Concepts coverage by existing ontologies
This section addresses the second research question RQ2. It has been observed that most of the identified ontologies have defined vocabulary for:
The coverage of the existing ontologies has been classified in Table 5.
Coverage of identified ontologies
Coverage of identified ontologies
It can be observed from Table 5 that location has been tackled by 67% of the identified ontologies. Different ontologies use different terms for location such as place or space as shown in Table 6. Ontology mapping and alignment can be done based on these terms in case one wants to integrate the ontologies.
Terms used to represent for location
To represent People, other terms have been used such as User in Campus Scheduling Ontology representing faculty and student. For Ubiquitous Learning Model, People represents an owner of a particular resource. In 3LConOnt, agents are used to represent a person, people, a group and an organization.
Services and Sensors have been modeled by most of the ontologies in IoT domain namely SmartLifeOntology, SmartCommunityOntology, OpenIoT Ontology, OntoIoT, IoT.est, IoT- Ontology, IoT-Lite, SemIoT and IoT-O.
SmartLifeOntology considers Service modeling based on elements such as environment, relevant sensor, criteria of the context, service behavior, related actuator and service procedure. The ontology focuses more on using context information from sensors and use the context information to execute tasks. IoT.est by Wang et al. (2012) also describes concepts related to IoT service modeling and service test to ensure the quality of services.
Modeling context and reasoning about the context are important aspects that should be covered by ontologies in pervasive and IoT environments. Context is being modeled by a number of ontologies namely SmartLifeOntology, OLOUD, SmartCommunityOntology, OpenIoT, IoT.est, SmartRoomOntology, IoT-Ontology, IoT-Lite, SemIoT and 3LConOnt. Some ontologies such as Campus Scheduling Ontology, SmartLifeOntology and Standard Ontology are intended for use with inference engines that allow for reasoning about the different context information provided by the sensors.
This section addresses the third research question RQ3. Most ontologies use standard languages like RDF/RDFS and OWL for ontology development as these languages provide a common way to describe information in a generic machine-interpretable form and thus support semantic level interoperability and information reusability (Kiljander et al., 2012). 89% of the identified ontologies have adopted OWL and 39% have adopted RDF/RDFS for ontology development as shown in Table 7. Domain experts prefer OWL as it is more expressive than RDF and RDFS, thus enabling more knowledge to be built (Chen et al., 2003). OWL is also designed as a standard. It additionally has the capability of support semantic interoperability and automated reasoning by automated processes (Gu et al., 2004).
Ontology language used for development
Ontology language used for development
Researchers have also emphasized the need for a standard ontology that shares and agrees on a common vocabulary and knowledge model for describing data (Barnaghi et al., 2012). Having such an ontology will improve interoperability between existing ontologies. Abdulrazak et al. (2010) claim that Standard Ontology is based on standards from World Health Organization (2001), Song and Lee (2008) and WIPO (2008). Kolchin et al. (2016) present a shared ontology that reuse existing standardized ontologies. According to Kolchin et al. (2016), IoT-Lite is not standardized. Ubiquitous Learning Model is based on the IEEE LOM (Learning Object Meta-Data) standard. SSN and SOSA are both W3C standards while IoT-O is based on the oneM2M standard. oneM2M is a cooperation between different standard development organizations (SDOs) in the world to define a world-wide standard for M2M/IoT communication (Kovacs et al., 2016). None of the identified ontologies have followed existing standards for smart communities such as ISO 37101, ISO 37120, ISO 37102, ISO 37123, ELOT 1457 standard (Greek) and BS 8904.
This section reports the findings for research question RQ4. Several authors (Barnaghi et al., 2012; Poslad et al., 2015; Serrano et al., 2015) have emphasized the need for lightweight ontologies for the context of IoT. The main argument put forward by the authors is that sensors in IoT systems lack resources such as memory and substantive computation and thus require lightweight and simple versions of models (Poslad et al., 2015; Su et al., 2010). Wang et al. (2012) also emphasize the fact that a lightweight ontology that balances expressiveness and inference complexity is more likely to be adopted and reused. Hepp (2007) argues that a trade-off exists between an ontology’s degree of detail and expressiveness and the achievable community size, that is, as the degree of detail and expressiveness of an ontology increases, the community size decreases. The reason that the author puts forward is that “the more detailed the ontology the fewer people will be willing to dedicate the resources for reviewing it prior to adopting it” (Hepp, 2007).
From the list of ontologies studied in this SLR, four of the eighteen identified ontologies, namely IoT.est, 3LConOnt, IoT-lite ontology and SOSA have addressed level of expressiveness in terms of lightweight semantics when developing their ontology. Table 8 shows a comparison of IoT-Lite and SOSA compared to SSN ontology (Arruda and Bulcão-Neto, 2019; Haller et al., 2019). In OWL DL, logical axioms have a significant influence on the overall complexity of the ontology (Arruda and Bulcão-Neto, 2019). The DL expressivity of IoT-Lite and SOSA is ALUI(D) and ALI(D) respectively as compared to SSN which is SRIQ. The DL SRIQ is a subset of SROIQ, which represents the W3C OWL DL Web Ontology language (Horrocks et al., 2005; Nortje et al., 2013).
Ontology comparison in terms of DL expressivity
Ontology comparison in terms of DL expressivity
In Description Logic,
(atomic concept)
(universal concept)
(bottom concept)
(atomic negation)
(intersection)
(value restriction)
(limited existential quantification).
Possible extensions for the restrictive language Role hierarchies Complex role hierarchies Cardinality restrictions Qualified cardinality restrictions Closed classes Inverse roles Datatypes Union of concepts Complement Full existential quantification
This section reports on RQ5, that is, the ontology development approach and methodologies adopted by the identified ontologies. As stated in Section 3.2.1, top-down and bottom-up are the two principal approaches for ontology development. In the top-down approach, generic modeling principles of a foundation ontology, also known as top-level or upper ontology can be adopted (Keet, 2011). Usage of a foundation ontology improves the overall quality of the ontology due to adoption of principled design decisions and enhances interoperability among ontologies aligned to the same foundation ontology (Keet, 2011). However, some domain ontology developers consider foundation ontologies as being too abstract and too expressive (Keet, 2011). It is also time-consuming to understand the foundation ontologies (Keet, 2011). According to Uschold (1996) and Ye et al. (2007), the bottom-up approach yields to the fullest possible description of the objects and provides very high level of detail. A bottom-up approach might, however, increase risk of inconsistencies, which in turn increase overall effort (Uschold, 1996). It may also be difficult to find commonalities between related concepts using the bottom-up approach (Uschold, 1996). Uschold (1996) recommends the use of a middle-out approach. Cristani and Cuel (2005) have classified methodologies such as Methontology, Uschold and King methodology and Ontology Development 101 as middle-out approaches for ontology development. Garcia et al. (2010) have classified Bernaras methodology and GM methodology as top-down approaches, TOVE and Methontology as middle-out approaches and iCAPTURer methodology as bottom-up approach. For DILIGENT methodology, the selection between top-down, bottom-up and middle-out approach is problem dependent while in NeOn methodology, no detail is provided for identification of concepts (Garcia et al., 2010).
Among the eighteen identified ontologies in the SLR, OLOUD has been developed using the Uschold and King methodology (Fleiner et al., 2017). According to Garcia et al. (2010), this approach starts with general concepts and may lead to ambiguity in the final product. 3LConOnt has considered recommendations from Methontology (Cabrera et al., 2019). This methodology has been chosen due to its evolving prototype life cycle, which allows moving back and forth from one state to another in case of missing, or wrong definitions. IoT-O has been developed using NeOn methodology. This methodology favors projects with domains that are not well understood and where requirements can change during the development process (Suárez-Figueroa et al., 2012). It also promotes the reuse of ontological and non-ontological resources. The methodologies adopted by the other identified ontologies have not been reported. The ontologies namely Top-level ontology, SmartLifeOntology, Smart Community Ontology, SemIoT, OpenIoT, IoT.est, IoT Ontology, IoT-Lite, OntoIoT, IoT-O and SOSA are generic ontologies. Generic ontologies tend to adopt a top-down or middle-out approach. One example is 3LConOnt where the upper-level ontology is built first and then adapted to domain ontologies such as smart parking, home, vehicle and restaurant (Cabrera et al., 2019). OntoIoT is another example where an upper ontology for IoT applications is developed first and then extended to application domain ontologies. From the SLR, it has been observed that none of the identified ontologies have followed a bottom-up approach for ontology development.
Limitations
Firstly, the review considers only papers written in the English language, which is the standard language in Science. Papers written in other languages were not considered for the review. Secondly, the review considers only papers from Google Scholar. A more significant data set from different data sources can be considered as future work. Thirdly, the review consists of papers indexed in Google Scholar by the end of December 2019. Papers indexed after that period were not considered.
Conclusion
Smart campuses make use of smart devices to communicate in smart environments over ubiquitous networks. The IoT is a new paradigm that supports the ubiquitous connectivity property for smart communities, including smart campuses. To ensure semantic interoperability in such environments, ontologies have been adopted. The paper presents a Systematic Literature Review that identifies eighteen ontologies for IoT-based smart community/campus. The ontologies are described and further classified in terms of domain, ontology language used, ontologies being reused, availability online and limitations. 83% of the identified ontologies have opted for ontology reuse. 72% of the ontologies are available online. Results of the review show that none of the existing ontologies have come up with a core semantic model for a smart campus that models the important domains such as smart learning, smart management, smart governance, smart room, smart health, smart library and smart parking amongst others and that enhances interoperability across these domains. Moreover, the coverage of the identified ontologies is classified in terms of People, Location, Resources, Agents, Events, Services, Context, Activity, Sensors/Actuators and Other elements. Among these concepts, Sensors/Actuators have been modeled by most of the ontologies, namely 72% of the identified ontologies. The SLR also reports on the usage of standards, level of expressiveness, ontology development approaches and methodologies adopted by the identified ontologies. All the ontologies have been developed using standard languages and OWL is the preferred language as compared to RDF/RDFS. None of the identified ontologies have adopted emerging standards for smart communities such as ISO 37101, ISO 37120, ISO 37102, ISO 37123, ELOT 1457 standard (Greek) and BS 8904. 22% of the ontologies have incorporated lightweight semantics that fit constrained environments. Only 17% of the ontologies have reported on the methodology adopted for ontology development. None of the identified ontologies have adopted the bottom-up approach for ontology development.
Footnotes
Acknowledgements
Special thanks go to Steve Ray, Freddy Priyatna and the anonymous reviewers for providing helpful comments to improve the paper.
