An intelligent expert system for academic advising utilizing fuzzy logic and semantic web technologies for smart cities education

Abstract

Students attending Higher Education Institutions (HEIs) are faced with a variety of complex decisions and procedures. To provide students with more sustained and personalized advising, many HEIs turn to online academic advising systems and tools as a way to minimize costs and streamline their advising services. However, in such systems, uncertainty in the learner’s parameters is a factor, which makes the decision-making process more difficult. Fuzzy logic, a multivalued logic similar to human thinking and interpretation, is highly suitable and applicable for developing knowledge-based academic advising systems that conserve the inherent fuzziness in learner models. In this paper, an innovative hybrid software infrastructure is presented which integrates expert system, fuzzy reasoning, and ontological tools to provide reliable recommendations to students for the next appropriate learning step. The software comprises a fuzzy logic component that determines the student’s interest degree for a specific academic choice accompanied by an ontological model and a conventional rule-based expert system for the composition of personalized learning pathways. In order for the system to recommend the next step of the learning pathway, the output of the fuzzy logic component together with the knowledge that is modeled as part of the multi-facet ontology and the machine perceptible academic advising guidelines expressed as semantic rules interoperate in a dynamic and seamless manner. The paper presents the key modeling artifacts of the proposed approach and the architecture of the implemented prototype system. During the case study, the developed system yielded satisfactory results in terms of overall inter-rater reliability and usefulness.

Full text

Vol.:(0123456789)
J. Comput. Educ.
https://doi.org/10.1007/s40692-022-00232-0
1 3
An intelligent expert system foracademic advising
utilizing fuzzy logic andsemantic web technologies
forsmart cities education
OmirosIatrellis1 · EvangelosStamatiadis2· NicholasSamaras1·
TheodorPanagiotakopoulos3,4· PanosFitsilis2
Received: 8 February 2022 / Revised: 9 May 2022 / Accepted: 24 May 2022
© Beijing Normal University 2022
Abstract Students attending Higher Education Institutions (HEIs) are faced with
a variety of complex decisions and procedures. To provide students with more sus-
tained and personalized advising, many HEIs turn to online academic advising sys-
tems and tools as a way to minimize costs and streamline their advising services.
However, in such systems, uncertainty in the learner’s parameters is a factor, which
makes the decision-making process more difficult. Fuzzy logic, a multivalued logic
similar to human thinking and interpretation, is highly suitable and applicable for
developing knowledge-based academic advising systems that conserve the inherent
fuzziness in learner models. In this paper, an innovative hybrid software infrastruc-
ture is presented which integrates expert system, fuzzy reasoning, and ontological
tools to provide reliable recommendations to students for the next appropriate learn-
ing step. The software comprises a fuzzy logic component that determines the stu-
dent’s interest degree for a specific academic choice accompanied by an ontological
model and a conventional rule-based expert system for the composition of personal-
ized learning pathways. In order for the system to recommend the next step of the
learning pathway, the output of the fuzzy logic component together with the knowl-
edge that is modeled as part of the multi-facet ontology and the machine perceptible
academic advising guidelines expressed as semantic rules interoperate in a dynamic
and seamless manner. The paper presents the key modeling artifacts of the proposed
approach and the architecture of the implemented prototype system. During the case
* Omiros Iatrellis
iatrellis@hotmail.com
1 Department ofDigital Systems, University ofThessaly, Larissa, Greece
2 Department ofBusiness Administration, University ofThessaly, Larissa, Greece
3 School ofScience andTechnology, Hellenic Open University, Patras, Greece
4 Business School, University ofNicosia, Nicosia, Cyprus

J. Comput. Educ.
1 3
study, the developed system yielded satisfactory results in terms of overall inter-rater
reliability and usefulness.
Keywords Academic advising· Fuzzy logic· Expert system· Semantic web
technologies· Higher education
Introduction
Academic advising is an important part of a cohesive strategy for HEIs to educate
and retain students. However, although most of the stakeholders in higher educa-
tion concede that academic advising is a key to student success, viable solutions
have eluded the institutions (Henderson & Goodridge, 2015). Academic advisory
services in many HEIs rely on personal meetings between students and counselors,
which allegedly have problems with inconsistencies among different advisors and
inefficient utilization of resources, since many advising sessions are spent answer-
ing recurrent and trivial questions without focusing on deeper conversations about
the student’s academic goals and plans (Aly etal., 2017). In this regard, the use of
computer-aided academic advising tools could have positive impacts on the above-
mentioned challenges.
Academic advising systems (AAS) and tools can be utilized for the provision of
consistent decision support for students by processing specific student parameters
derived from the learner model in order to produce personalized advice (Iatrel-
lis etal., 2017). In contrast to the traditional course management and registration
systems offered by many HEIs, an AAS provides students with a higher level of
informed recommendations, which can be considered as an emulation of an aca-
demic advisor’s role, rather than a registration assistant’s role. Traditional course
management and registration systems focus mainly on program corequisites, prereq-
uisite rules, and other registration restrictions. On the other hand, an AAS requires
reasoning over the current information and knowledge at each decision node of the
education plan, taking into account both the available academic options and cir-
cumstances inside a HEI and the uncertainties, subjectivities, implicit, and vague
information, which is often hidden in learner parameters. Since the fuzzy inference
approach is suitable at modeling human knowledge, it is often used to deal with
those types of problems in the literature. The fuzzy methodologies can broadly be
grouped into two main types: linguistic fuzzy systems (Mamdani) and precise fuzzy
systems (Takagi–Sugeno–Kang) (Abduldaim & Sabri, 2019). Both of them contain
the expert knowledge of fuzzy logic with their own problem-solving capability;
however, Mamdani approach is widely used in particular for decision support appli-
cations (Prasad etal., 2017) such as the academic advising systems.
This paper proposes a fuzzy hybrid academic advising system named EDU-
C8EU (EDUCATE EUROPE) to provide the recommendation of the most appro-
priate learning step for each student. The architecture of EDUC8EU comprises two
core subsystems where the first one receives the personal parameters and charac-
teristics of the learner that affect the decision-making process and the second one

1 3
J. Comput. Educ.
is a conventional rule-based expert system. Initially, the fuzzy inputs of the learner
model are processed to determine the interest degree for a learning pathway accord-
ing to student’s individual characteristics, needs, and requirements, not just in terms
of academic goals, but also in terms of planning, costs, etc. The output of the first
subsystem is entered to the second one, which is based on an expert system in order
to decide upon the next learning step to be proposed. The technological backbone
of EDUC8EU utilizes semantic web technologies to achieve a holistic conceptual-
ization of the domain of educational provision, in order to be further utilized for
the formalization of the academic advising rules. The core of the semantic model
is based on an ontology that provides an integral conceptual model covering all the
involved knowledge streams for the consistent representation of the specific domain
as well as for the implementation of the rule repository.
The proposed platform has been developed in the context of the INVEST4EX-
CELLENCE European Universities H2020 program,1 which aims at establishing
transnational university alliances for developing joint and innovative education and
research study programs and curricula, as well as for the implementation of multi-
lingual learning, blended and work-based learning, and European mobilities. The
EDUC8EU platform will provide support to students at all three study cycles—
bachelor, master, and doctoral, together with the living labs, Vocational Education
and Training (VET) certifications, Massive Open Online Courses (MOOCs), and
other extracurricular educational activities. Consequently, the establishment of an
effective computer-aided academic advising solution that can actively guide the stu-
dents and react to changes would be of major importance for the participating uni-
versities in the alliance since the offered learning pathways will encompass a wide
variety of educational options in diverse settings, languages, and disciplines.
In the rest of the paper, we review the related work in “Literature review” section
and present our learner model. Section “The EDUC8EU approach” presents the key
concepts of the implemented approach, while “EDUC8EU technical architecture”
section overviews the technical architecture of the implemented EDUC8EU inte-
grated software environment. Section “Case study” presents our case study and “Our
contribution” section describes our contribution in the specific domain. Finally, in
“Conclusions and future work” section, we provide conclusions and discuss future
research directions.
Literature review
The learner’s profile has been gaining an increasing attention in education research
discipline. There is a growing number of academic studies that highlight the role
played by—what constitutes—a learner model, in Academic Advising Systems.
This paper crawled through the following scientific databases to retrieve the existing
research supporting that role: IEEE, Science Direct, and Springer Link.
1 https:// www. inves t4exc ellen ce. eu.

J. Comput. Educ.
1 3
The purpose of this search is to find studies that pinpoint the traits and character-
istics a learner model comprises, which may be used in recommending more effec-
tive learning objects and pathways, thus elevating the accuracy of academic advising
systems. The key words that were used to perform this search include the following:
“learner profile and recommender systems,” “learner profile and personalized learn-
ing,” and “learner profile and learning pathway.” The outcome of this search process
yielded 32 research papers related to this study. The selected papers are categorized
by publication year and scientific database (Park etal., 2012).
Classification bypublication year
Figure1 depicts the classification of the research papers by year of publication. The
time period from 2014 to 2020 is shown to be more productive with regard to papers
related to learners’ profiles and their association with recommender system.
Classification byscientific database
The research papers were published in three different scientific databases. The
classification by scientific database is shown in Table1 and Fig.2. Springer Link
yielded most of them, 12 out of 32, which corresponds to a quota of 37,50%, fol-
lowed closely by Science Direct in second place with 11 papers out of 32 at a quota
Fig. 1 Classification of research papers by year of publication

1 3
J. Comput. Educ.
of 34,38%. Lastly, third place goes to IEEE with 9 out of 32 papers at a quota of
28,13%.
Most of the research papers showcase the incorporation of a learner’s traits, fea-
tures, behavior, and preferences into recommender learning systems to improve the
recommendation process and the personalization of learning pathways.
Related work
Examining the learner’s traits, features, and preferences, such as cognitive state and
learning style, effort, and behavior (Troussas etal., 2020), and inducting them into
the recommendation process, is an essential ingredient of successful recommenda-
tions in academic advising systems. The importance of the learner’s profile is also
highlighted in this extensive, state-of-the-art review on 82 recommender systems
(RS) from 35 countries (Drachsler etal., 2015), which analyzes a number of papers
(Bielikovà etal., 2014; Casali etal., 2011; Kaklauskas etal., 2013; Martín & Carro,
2009; Santos etal., 2014; Schoefegger etal., 2010), outlining the significant aspects
Table 1 Classification of
research papers by scientific
database where they were
published
Scientific database Amount Quota (%)
IEEE 9 28,13
Springer link 12 37,50
Science direct 11 34,38
Total 32 100
Fig. 2 Classification of research papers by scientific database

J. Comput. Educ.
1 3
of Technology-Enhanced Learning that include also the user’s prior knowledge,
skills, and abilities in the process. Since grades, scores, exams, GPA, SATs, etc.,
offer a way to quantitatively measure/capture these different learner features, past
performance gets promoted as a key concept to be considered when devising frame-
works and methods for building intelligent recommender systems (Aguilar etal.,
2017; Thanh-Nhan etal., 2016), recommending personalized learning pathways (Xu
etal., 2016), applying predictive analytics to find out the suitable courses for a stu-
dent’s admission to a college (Upendran etal., 2016), or even establishing a predic-
tion model for the potential of a student opting for early admission in a university
(Chen etal., 2014). This key concept alone cannot encapsulate the learner acquired
knowledge and skills, as it needs to be complemented by their Perceived Difficulty
of the Learning Pathway and/or Object. Also, variations like advanced learner or
beginner learner, as well as time constraints (limited time available for studying)
can add up to the difficulty factor (Drachsler etal., 2015). Therefore, matching up
learners’ abilities with the difficulty levels of a recommended learning object and/or
pathway makes the difference for academic advising systems in avoiding disorienta-
tion or cognitive overload during (Wigfield & Cambria, 2010) the learning process
(Chen & Duh, 2008; Chen etal., 2005).
A learner’s motivation plays a significant role in their learning behavior and cog-
nitive engagement (Essa, 2016; Pintrich, 2003). This analysis breaks the value of
motivation down to goal orientation and task value, the latter comprising three com-
ponents, the importance of the task for the learner, their personal interest for the task
and how they perceive the task for future goals, e.g., pursuing a career (Anderman
etal., 2012; Eccles, 1983). In other words, the learner’s career goals, their personal
interests, and the manner they perceive the relevance of the selected learning objects
to these goals, combined with the learner’s prior knowledge, skills, and abilities and
an estimation of the workload needed for accomplishing these courses (Farzan &
Brusilovsky, 2006), would result in an amalgamation of fundamental criteria that
an academic advising system should incorporate into its recommendation process
in order to showcase the value of the course/pathway to the learner (Garrido &
Morales, 2014; Wigfield & Cambria, 2010). Career goals prove of high value to the
learner, providing a powerful incentive to opt for a learning object, even in cases
that this course appears difficult to accomplish (Upendran etal., 2016). The same
principle applies also to mature learners that participate in Higher Education, whose
motivational dynamics is strongly associated with extrinsic and intrinsic factors that
stem from sources like career progression, professional improvement, development
of professional and personal competences, economic advantage, new career oppor-
tunities, etc. (Duarte etal., 2018). The higher value a learner perceives for a task, the
higher the commitment and learning outcome (Du Boulay etal., 2010; Ryan & Deci,
2000). Consequently, this adds Perceived Career Opportunities into the blueprint for
modeling a learner’s profile.
Social networking, being an immense, omni-present phenomenon in everyday
life’s various aspects (personal and professional), is a significant factor to be taken
into account in recommender systems for learning objects (Dias & Wives, 2019).
Ubiquitous e-learning systems that blend real-world learning with virtual, per-
sonal, and shared space consider the learning process as a social transfer process

1 3
J. Comput. Educ.
of knowledge (Li etal., 2004). When opting for a learning object, “what to learn”
proves as important as “with whom to learn” or “where” this learning object was
discussed (e.g., forum), highlighting the importance of social networking in the
creation and selection of knowledge. According to (Dias & Wives, 2019) when a
recommender system combines a learner’s choices with “with whom to learn,” it
generates results with the highest usage prediction accuracy. Other studies high-
light the importance of trust that a learner may extend toward certain peers of theirs,
regarding what learning objects they have selected or rated. Trust and reliability
among peers is a force to be reckoned with; recommender systems that incorporate
this parameter can turn out to be effective and efficient (Carchiolo etal., 2010). The
learner’s profile needs to be extended to include additional information about their
opinions, critiques and relationship with other learners (e.g., their friend groups) in
order to increase the knowledge base of an academic advising system to perform
more intelligent and precise recommendations (Aguilar et al., 2017). Therefore,
Friends & Peers in Learning Pathway gets promoted as a complementing concept
into the Learner Model.
This social facet of the learning recommendation process is manifested
through social signals, interests, and preferences of the learners, which, in turn,
adhere to the process as enablers for improving the accuracy of recommendation.
These signals can derive from learners’ interactions with other learners, during
the learning process, by using forums, chats, or private messaging, as it happens
with Moodle. They can also derive when learners rate, comment on, share, or
bookmark learning objects in social learning networks, e.g., MERLOT. Besides
these explicit manifestations, implicit feedback can be extracted as informational
behavior, through the learner’s search actions and use of information in the sys-
tem (keyword searches, selection of results, clicks, page views, and saves, etc.)
(Takano & Li, 2009), or as social media behavior through user bookmarks in
social networks (Durao & Dolog, 2009). By exploring other learners’ behavior
during the learning process and combining it with these social signals, academic
advising systems can alleviate the burden of excessive information on learning
objects, filtering, and recommending the most relevant courses to the interested
learner through a user-centric approach (Dias & Wives, 2019). The aforemen-
tioned recommending approach introduces what can be characterized as the Rep-
utation of Learning Pathway concept when modeling the learner. Building a rep-
utation for a learning object in a recommender system can be achieved through
various methods and techniques. For instance, the system may form a neighbor-
hood of learners considering their profiles (prior knowledge, performance, learn-
ing styles) to discover associations among learning objects and identify thereafter
the useful ones that have been accessed by similar learners, thus constructing a
personalized recommendation list of courses upon the visit history of the mem-
bers in that neighborhood (Imran etal., 2016). Other recommender systems cal-
culate an item’s reputation by aggregating ratings provided by multiple users,
hence reflecting the formulated opinion of a community on that item (Abdel-
Hafez etal., 2014). Hybrid approaches are also used to match up learner pro-
filing—based on their interests, preferences, and historical access records—with
learning objects—based on multi-dimensional attributes of the courses (Salehi &

J. Comput. Educ.
1 3
Kmalabadi, 2012). Inducting reputation metadata that record the learners’ opin-
ion and ratings about learning objects, into recommender learning systems, trans-
forms them into context-aware recommender systems (Adomavicius & Tuzhilin,
2015) and can improve the learner’s satisfaction on the recommendation results
(Kerkiri etal., 2008). Lastly, the “Panorama of Recommender Systems to Sup-
port Learning” (Drachsler etal., 2015) review in the field of technology enhanced
learning, records systems that “guide learners to relevant resources that have
been previously found as valuable by other learners”; “consider user evaluations
of learning resources and propagate them to users with similar tastes”; “manage
learners’ properties based on learning styles and reputation metadata”; and con-
sider “multidimensional ratings provided by the users on learning resources.”
Wrapping up the learner model, there is an extrinsic factor related to an attrib-
ute of a learning object that can affect their decision-making process in following
that learning path: its financial cost. Although this is subject to a person’s social
and financial status, hence varies from learner to learner, following a higher educa-
tion course is an once-in-their-lives choice for most people, because such a decision
entails investing in time and money (Díaz-Díaz & Galpin, 2020) Furthermore, if that
decision proves wrong, it may have significant ramifications to that person’s life.
Recommender systems from other domains (e.g., financial investments, car buying
/ renting industry) consider the true monetary value and final cost of the item when
selecting the most suitable or applicable recommendation approach (Ricci et al.,
2015). The same consideration fed this next fuzzy expert system for academic advis-
ing with the input variable “Cost of Course,” to be included in the recommendation
process of a learning object (Aly etal., 2017). Thus, the parameter Cost of Learning
Pathway is the last key concept to complement the learner model of this research.
Table 2 Parameters pertaining a learner model
Parameter Deducing literature
Past performance Aguilar etal. (2017), Thanh-Nhan etal. (2016), Xu etal.
(2016), Upendran etal. (2016), Chen etal. (2014)
Perceived difficulty of learning pathway/object Drachsler etal. (2015), Chen etal. (2005), Chen and
Duh (2008)
Perceived career opportunities Essa (2016), Pintrich (2003), Eccles (1983), Anderman
etal. (2012), Farzan and Brusilovsky (2006), Garrido
and Morales (2014), Wigfield and Cambria (2010),
Upendran etal. (2016), Duarte etal. (2018), Du Bou-
lay etal. (2010), Ryan and Deci (2000)
Friends & peers in learning pathway Dias and Wives (2019), Li etal. (2004), Carchiolo etal.
(2010), Aguilar etal. (2017)
Reputation of learning object Takano and Li (2009), Durao and Dolog (2009), Dias
and Wives (2019), Imran etal. (2016), Abdel-Hafez
etal. (2014), Salehi and Kmalabadi (2012), Adoma-
vicius and Tuzhilin (2015), Kerkiri etal. (2008),
Drachsler etal. (2015)
Cost of learning pathway Díaz-Díaz and Galpin (2020), Ricci etal. (2015), Aly
etal. (2017)

1 3
J. Comput. Educ.
Conclusively, Table2 summarizes the main concepts that constitute the learner
model of the proposed EDUC8EU system, as defined according to this study’s lit-
erature review:
These parameters in the EDUC8EU learner model constitute fuzzy linguistic
variables that are dynamic estimators of student’s interest for a specific learning
pathway.
The EDUC8EU approach
EDUC8EU conceptual architecture
As illustrated in Fig.3, the conceptual architecture of EDUC8EU consists of three
main components: (a) the semantic infrastructure, (b) the Fuzzy Logic Controller
(FLC), and (c) the conventional rule-based expert system.
The core of the semantic infrastructure is a multi-facet ontology, which encap-
sulates the needed knowledge streams for the representation of all the participat-
ing entities in the academic advising process and leverages the utilization of seman-
tic rules that model the academic advising knowledge and experience acquired by
domain experts. The semantic infrastructure is utilized for (a) the modeling of the
entities of the learning pathways in a clear and unambiguous way, (b) the academic
advising recommendations formalization, and (c) the knowledge discovery through
the dynamic generation of new facts and conclusions from the rule engine.
The aim of designing the FLC subsystem is to process the parameters affecting
decision-making workflow and to determine student’s interest for a specific learn-
ing pathway. Notions that are included in the learner model like “perceived career
opportunities” or “past performance” are not crisp concepts. Learners implicitly
interpret them as approximate concepts, which can assume different degrees of truth.
Using the FLC subsystem, the classification of these learner parameters becomes a
graded matter of possibility instead of a Boolean assignment. Thus, to determine
the student’s interest for an academic choice, several factors are examined by the
FLC and the interactions between them as well. The FLC subsystem involves all
pieces of the fuzzy inference process (membership functions, fuzzy logic operators,
and if–then rules) that determines the student’s interest corresponding to fuzzified
learner inputs.
Fig. 3 The conceptual architecture of EDUC8EU

J. Comput. Educ.
1 3
The second subsystem of the EDUC8EU software infrastructure incorporates a
conventional rule-based expert system that receives the output of the FLC, combines
it with other knowledge streams enclosed in the semantic model, and executes the
crisp semantic rules in order to generate the final recommendation for the appro-
priate next step of the academic plan. These streams of knowledge reflect upon a
different dimension of a learning pathway related to the operational and organiza-
tional aspects in order to cover various possible learner needs (such as scholarships,
hands-on labs, placements among others) and program requirements (such as admis-
sion process, course prerequisites, tuition fees, or language proficiency). For exam-
ple, the “Admission Process” requirement refers to the administrative and functional
procedure a learner has to go through in order to be admitted (exams, interviews
or screening process), which is an essential criterion in learning pathway selection.
An important component of the EDUC8EU approach is the assumption that these
organizational related concepts be crisp input variables, which can be processed by
a conventional expert system and can be implemented as class instances inside the
ontology in order to provide extensibility and flexibility to the model of the aca-
demic advising rules. This aligns the EDUC8EU approach with the hybrid develop-
ment strategy in intelligent systems as identified in Medsker (1995) and more spe-
cifically with the transformational model. Thus, in our approach, a tight integration
of three forms of reasoning is introduced namely (a) the ontological, (b) the fuzzy-
like, and (c) the classical rule-based, resulting in an intelligent hybrid system that
incorporates a more "open world" view of the input and output variables for the case
of an AAS. This is achieved because the synergy between the abovementioned tech-
niques allows, on the one hand, the representation of the whole “universe” of the
participating entities and their relationships in the decision-making process, and, on
the other hand, attempts to address the vagueness in academic advising scenarios.
Another benefit of implementing the EDUC8EU architecture as a hybrid transfor-
mational model is that development can occur in the most appropriate subsystem
and therefore a faster implementation and less maintenance is achieved (Medsker,
1995). In this regard, the modeling of the crisp semantic rules is grounded in an
existing semantic model utilized by another software platform that already operates
at the University of Thessaly, in Greece, and handles the dynamic orchestration of
educational processes of the HEI (Iatrellis etal., 2019a). This was a technical deci-
sion that promotes interoperability and maintainability through the establishment of
a common knowledge base, which in turn can lead to increased inference validity
and enhanced ontology enrichment. In this way, the utilized semantic model consti-
tutes a totally dynamic and evolving knowledge space, which can be considered as
an asset for the EDUC8EU environment.
Fuzzy logic controller
The fuzzy inference engine of the present system uses Mamdani inferencing, which
is widely accepted for describing expert knowledge in a more intuitive and human-
like manner (Molina-Solana etal., 2017). Following the Mamdani-style inference
approach, the FLC inference process is performed in four steps (Irfan etal., 2019):

1 3
J. Comput. Educ.
(a) Fuzzification of the learner variables: determining the degree to which these
inputs belong to each of the appropriate fuzzy sets
(b) Academic advising rule evaluation (inference): for each rule, applying the result
of the antecedent evaluation to the MF
(c) Aggregation of the rule outputs (composition): unification of the outputs of all
rules into a single fuzzy set
(d) Defuzzification: Calculation of the final output of the FLC in a crisp format
The EDUC8EU learner model was built based on existing literature as described
in “Literature review” section. Thus, the proposed system uses six main inputs and
produces a single output. Table3 lists these inputs that represent fuzzy linguistic
variables alongside with their corresponding values.
EDUC8EU ontology
The first step toward automatic computer-based academic advising recommenda-
tion is knowledge representation. In our case, it involves the abstraction of domain-
specific knowledge in terms of concepts that model the academic advising aspect
of learning pathway, its organizational dimension, and the important learner param-
eters affecting decision-making process.
Table 3 Learner input
parameters and linguistic values
pertaining the learner’s interest
Parameter Linguistic variable
Past performance Very good
Good
Fair
Poor
Perceived difficulty of learning pathway Hard
Average
Easy
Perceived career opportunities Many
Some
Few
Friends & peers in learning pathway High
Average
Low
Reputation of learning pathway High
Moderate
Low
Cost of learning pathway High
Moderate
Low

J. Comput. Educ.
1 3
According to Fig.4, the ontological model, the core of the semantic model is
defined as a combination of concepts stemming from four knowledge streams,
namely: 1) the Learning Pathway, 2) the Learner, 3) the Business and organizational
dimension, and 4) the Quality Assurance. The first part of the implemented ontology
contains the semantics to be utilized by the software environment for the recommen-
dation of the appropriate learning steps. The respective subdomain is combined with
the building blocks of the EDUC8EU learner model, which describes the main actor
of the academic advising process. In addition to the Interests and Requirements
concepts of the Learner model, the RIASEC acronym is derived from the research
of Dr. John Holland and provides a preliminary way to identify learning pathways
that might match students’ personality. Holland’s theory of career choice is widely
accepted (Nauta, 2010) and forms the basis of many popular and heavily researched
career inventories including the free online database O*NET (Occupational Infor-
mation Network) maintained by the US Department of Labor ETA.2 The utilized
semantic model, besides the learner and learning pathway subdomains, covers the
business and organizational dimensions of an HEI both on an intra- or inter-organ-
izational level by incorporating a well-distinguished business modeling ontology,
namely the Resource-Event-Agent ontology (McCarthy, 2003). The abovemen-
tioned parts are combined with a “Quality assurance” module by encapsulating a
set of related quality assurance concepts derived from the revised 2020 European
Foundation for Quality Management business excellence model.3 The four domains
that are modeled and interfaced as part of the ontology are described in detail in
Fig. 4 EDUC8EU ontology abstract diagram
2 https:// www. oneto nline. org/.
3 https:// efqm. org

1 3
J. Comput. Educ.
(Iatrellis etal., 2019b) and facilitate the semantic interoperability among EDUC8EU
and other third-party applications that already exist in the higher education sector.
EDUC8EU technical architecture
The EDUC8EU technical architecture is based on a 3-tier web architecture model
to maximize flexibility, adaptability, and stability (Fig.5). Moreover, the specific
architecture enables reusability, integration, and composition of components at dif-
ferent levels. During each execution cycle, the knowledge inside the ontology, the
student data, and the rule-set are interoperating in order to offer personalized aca-
demic advising guidance and achieve the most optimized outcomes. The core of the
Fig. 5 EDUC8EU technical architecture

J. Comput. Educ.
1 3
business layer utilizes the DROOLS expert system technological solution4 to pro-
vide a unified and integrated platform for rule-based event processing.
Data access layer
The specific layer is responsible for the storage and maintenance of the EDUC8EU
data and information and it is implemented based on MySQL platform, which is
considered as one of the most popular relational database management system
(RDBMS), licensed under the GNU GPL. The Data access layer also includes the
“CRUD Data Access and Operations” Layer and provides the appropriate mecha-
nisms for Creating, Retrieving, Updating, and Deleting data records. Moreover,
the Data Access layer mechanisms are based on and include the basic principles
and properties of the “Hibernate Framework,” which provides a “classical” object-
relational mapping between Java classes and database tables, permitting the EDU-
C8EU developer to access instances of such classes (actually stored in a database)
as if they were true Java objects. Finally, “Data Access” layer is responsible for the
storage, management, and creation of all the necessary modeling components and
meta-models used by the upper layers of the EDUC8EU integrated platform. More
specifically, it encapsulates the following components:
(a) the XSD repository which encloses the complete set of standardized and elec-
tronic templates of all types of the predefined learner parameters and linguistic
values required during the reasoning process. The primary use of these stand-
ardized formats concerns the dynamic creation of the corresponding web form
elements that are presented to the end-users.
(b) the End-User Accounts Directory, which includes detailed information about
end-user accounts combined with their profiles based on their expertise and their
position on the organization chart of the HEI. This particular directory provides
the privileges concerning the access and interaction with the design tools in
order to update the knowledge stored inside the semantic model.
(c) the Membership Functions Data, which is part of the database, and stores the
necessary values that define how each point in the input space is mapped to a
degree of membership between 0 and 1. As elaborated in the ensuing subsec-
tion, during the design mode, the domain knowledge experts can perform all the
necessary reconfigurations using the EDUC8EU backend, which provides the
benefits of a user-friendly graphical interface for the tuning of the MFs.
(d) the Student Academic Data, which is also part of the EDUC8EU database, and
includes learner’s parameters and academic data that in turn enhance the learn-
er’s profile. This part of the database is updated at the end of each reasoning
process and after the data entry of the respective student information.
4 https:// drools. org.

1 3
J. Comput. Educ.
Business logic layer
The specific layer is responsible for processing the business logic of EDUC8EU
components (without taking into account the presentation and graphical user inter-
face requirements) and for accessing the data layer to retrieve, modify, and delete
data to and from the RDBMS. JBoss Application Server is used as the application
engine of choice because it provides an integrated platform for development and
includes the following:
Semantic sub-layer
During the design mode, the end-user interacts mainly with the semantic layer in
order to ensure the constant update and evolution of the academic advising, opera-
tional, organizational, and quality-related knowledge that is stored inside the seman-
tic model. The domain experts (academic advisors) are responsible for the main-
tenance of the ontology and rule-set repository to be utilized during the reasoning
process by the expert system. In order to facilitate the continuous maintenance of
the semantic rule base in an integrated way by the domain knowledge experts, a
graphical rule generator interface was implemented, which offers a better viewing
experience for rules and rulesets to facilitate the understanding, knowledge, and
manipulation of them.
Fuzzy inference
The FLC subsystem is partially based on an open source Mamdani-style infer-
ence engine.5 Server-side Java Server Pages technology is used to extend the
Fig. 6 Reasoning process
5 https:// github. com/ marci ngol1/ fuzzy.

J. Comput. Educ.
1 3
capabilities of the application by adding dynamic server-side web scripting fea-
tures and seamless integration with the database.
As it is depicted in Fig.6, within the FLC subsystem, the end-user has to
enter values for all the inputs defined in the learner model. The subsystem
will respond with the student’s interest degree to be forwarded to the second
subsystem.
Fuzzifier Choosing an efficient MF in fuzzy algorithms plays a vital role to
achieve the appropriate results. In order to facilitate the tuning of the MFs, an
online membership editor UI was implemented. The tool lets the end user display
and edit all of the MFs associated with all of the input and output variables for
the EDUC8EU system. Figure7 shows the graphical representations of the “past
performance” factor as an example. Both the choice of triangular and trapezoi-
dal shape are supported by the implemented tool because of the simplicity of
specification and the satisfying results (Fallahnejad & Moshiri, 2014). More spe-
cifically, the triangular function can be defined in the tool by a lower limit a, an
upper limit b, and a value m inserted twice in the corresponding textbox, where
a < m < b and is calculated by the following formula:
Fig. 7 MF of “Past performance” variable

1 3
J. Comput. Educ.
The trapezoidal function can be defined in the tool by a lower limit a, an upper
limit d, a lower support limit b, and an upper support limit c, where a < b < c < d
and is calculated by the following formula:
The output variable, which represents the student’s interest for the specific
pathway, is defined by five linguistic degrees as “Very Low,” “Low,” “Moder-
ate,” “High,” and “Very High.” Once the result is processed, the student’s inter-
est degree is forwarded to the second subsystem in order to be fed into the rule
engine to lay out the pathway.
Fuzzy rules Based on the fuzzy values obtained from the fuzzifier, the rules were
acquired from a group of academic advisors, incorporating their knowledge and
experience. The academic advisors were tenured faculty members of University of
Thessaly in Greece whose experience ranged from 12 to 25year and whose profes-
sional tasks ranged from being solely an educator to being an educator, researcher,
and department chair simultaneously. All of 24 rules were approved by 100% of
the domain experts participated in the research. The rules are of the structure (IF
antecedent(s) THEN consequent) where the antecedents are the learner’s factors
and the consequent is the student’s interest degree. A sample of the rules used is
the following (see Table4):
(1)
𝜇
A(x)=
⎧
⎪
⎪
⎨
⎪
⎪
⎩
0, x≤𝛼
x−α
m−α ,a<x≤
m
b−x
b−m ,m<x<
b
0, x≥b
(2)
𝜇
A(x)=
⎧
⎪
⎪
⎨
⎪
⎪
⎩
0, (x<a)or(x>d
)
x−α
b−α ,a≤x≤b
1, b≤x≤c
d−x
d−c,c≤x≤d
Table 4 Sample of Fuzzy rules used
R1 If [Reputation of Learning Object is low] then [Interest is low]
R13 If [Friends & Peers in Learning Pathway is High] and [Perceived Difficulty of Learning Pathway/
Object is average] then [Interest is moderate]
R17 If [Reputation of Learning Object is average] and [Past Performance is veryGood] then [Interest is
high]
R19 If [Past Performance is veryGood] and [Friends & Peers in Learning Pathway is many] then [Inter-
est is veryHigh]
R24 If [Cost of Learning Pathway is high] and [Perceived Difficulty of Learning Pathway/Object is
easy] then [Interest is veryLow]

J. Comput. Educ.
1 3
Defuzzification The defuzzifier component utilizes the centroid method, which is
probably the most popular and useful defuzzification technique. It finds the point
where a vertical line would slice the aggregate set into two equal masses. Mathemati-
cally, this center of gravity (COG) is calculated as follows:
Recommendation module
To describe the rules for the recommendation module, we employed the Semantic
Web Rule Language (SWRL) (Horrocks etal., 2010). SWRL uses the rule syntax
“Antecedent → Consequent” to represent semantic relationships. Both anteced-
ent and consequent are formulated as conjunctions of atoms written a1 ∧ … ∧ an.
Variables are denoted using the standard convention of prefixing them with a ques-
tion mark (e.g., ?V). Furthermore, SWRL provides many useful built-ins to support
comparisons, math, date, or string functions. This module comprises:
The crisp SWRL rules Which model the knowledge and experience of the domain
expert. The SWRL rules can perform complex reasoning and calculations by includ-
ing i) the output of the first stage, ii) other inputs derived from the EDUC8EU seman-
tic model to be used in the reasoning process, and iii) the final recommendations,
which are the output of the EDUC8EU system. An indicative set of the crisp SWRL
rules for the EDUC8EU prototype is presented shortly:
Rule A
1 Learner(?L) ^ Process(?P) ^
2 hasRIASEC(?L, IRC) ^
2 LearningState(?S) ^ hasLearningState(?L,? S) ^ hasInput(?P,? S) ^
3 hasName(?P, “orientation”) ^
4 Interests(?I) ^
5 hasInterest(?I, computerNetworks) ^ hasInterestDegree(?I, high)
6 Curriculum(?C) ^ hasType(?C, handsOnLabs) ^
7 hasRequirement(?P1, admissionProcess) ^
8 hasSatisfactory(?P1, engishSkills) ^
9 hasSatisfactory(?P1, computerBasics) ^ Process(?P1)
10 →
11 hasName(?P1, “CCENT”)
Rule A describes the following situation: if the student has a high level of interest
in “Computer Networks” (line 5), wishes to gain hands-on experience (line 6) and
there is an estimation of “IRC” (Investigative, Realistic and Conventional) as his
Holland/RIASEC code pattern (line 2) then the next step of the procedure foresees
(3)
COG
=∫
b
a𝜇A(x)
xdx
∫
b
a
𝜇A(x)dx

1 3
J. Comput. Educ.
the registration to the CCENT certification (line 11), which requires an initial admis-
sion screening process (line 7). Good knowledge of English and basic computer
skills are also important prerequisites according to rule A (lines 8, 9).
Rule B
1 Learner(?L)^Process(?P) ^ LearningState(?S) ^
2 hasInput(?P,? S) ^ hasName(?P, "CCENT") ^
3 hasSatisfactory (?S, CCENTexams)
4 Interests(?I) ^ hasInterest(?I, ICTSecurity) ^
5 hasInterestDegree(?I, veryHigh)
6 hasRequirement(?P1, hasTuitionFees) ^
7 Resource(?C) ^ hasQA(?C, MinimumEnrolledStudents) ^
8 Process(?P1)
9 →
10 hasPart(?P1,? P) ^ hasName(?P1, “CCNASecurity”)
Rule B describes the decision concerning the CCNA pathway selection per-
formed once the CCENT cycle is completed. The first hypothesis of the specific rule
is that the student has a “very high” level of interest in the “ICT Security” field
(lines 4, 5). So, if the student is considered to have satisfactorily completed the
CCENT exams (line 3) then the next step of the pathway foresees the enrollment
in the “CCNASecurity” pathway (line 10) provided that the specific track has more
than a prescribed minimum number of students enrolled to make it viable. It has to
be noted that the “minimumEnrolledStudents” indicator is derived from the quality
assurance ontological module and is instantiated in Rule B in order to demonstrate
the flexibility the current approach provides.
The “SWRL toDROOLS” translator Which takes as input the SWRL rules and
translates them into a DROOLS world using a collection of available software com-
ponents and Java-based APIs.6 Initially, the appropriate SWRL rule-set is selected by
the specific component and transformed into DROOLS compatible format so as to be
further processed by the rule engine.
The DROOLS rule-based expert system Which is one of the key components of
the business logic for the personalized recommendation of learning pathways. The
DROOLS rule-based expert system fires the semantic rules in order to reason upon
them and produce the result in JSON format. The JSON file generated is defined as an
asset for the EDUC8EU platform so as to be further utilized for the composition of a
detailed report containing well-founded recommendations tightly integrated into the
available learning pathways (Fig.8). An important feature of the specific component
is that in any execution cycle, a negative recommendation is also possible to be cre-
ated if the learning pathway is considered not appropriate for the learner. Thus, when-
6 https:// github. com/ prote gepro ject/ swrla pi/ wiki.

J. Comput. Educ.
1 3
ever the conditions specified in the antecedents of the semantic rules are not fulfilled,
no rule is triggered and hence a negative recommendation is generated by the EDU-
C8EU system. Moreover, once the rule-set execution is completed, the expert system
produces a feedback message (Fig.5), which contains new DROOLS world facts that
can update respectively the knowledge inside the EDUC8EU ontology and the rule
base ensuring its perpetual maintenance. Thus, this component provides exception
detection functions with semantic rules in order to handle the recommendation of the
most appropriate learning step for each student as well as the knowledge evolution.
Presentation layer
The Apache Tomcat7 software will play the role of the web server that interacts with
Java servlets and JSPs, thus enabling transparent access to the platform through
Fig. 8 EDUC8EU recommendation
7 https:// tomcat. apache. org/

1 3
J. Comput. Educ.
simple web browsers. Moreover, the use of XML data and XSLT transformations
serves the adaptation of web pages according to the role and the respective access
rights of the specific user. The presentation layer allows the integration and presen-
tation of several client-side JavaScript components based on jQuery for various tools
and applications of the EDUC8EU software environment. Finally, the specific layer
serves for the triggering of applications, tools, and services of the integrated EDU-
C8EU software environment.
Case study
The performance and completeness of the implemented prototype were tested dur-
ing the selection of a set of appropriate academic advising recommendations regard-
ing the MOOC for Smart City professionals offered through Moodle’s Open Edu-
cation platform.8 The specific MOOC was developed under the framework of the
Smart DevOps project to offer high quality educational course that will enable the
students to develop and acquire essential competencies needed to tackle the chal-
lenges of managing and evolving of smart cities. Students who completed all mod-
ules successfully were qualified for the second round of specialized training, which
included three different learning pathways leading to the certification of three smart
city job profiles: (1) Smart city Planner, (2) Smart City IT manager, or (3) Smart
City IT office job profile (Fig.9) (Kaufmann etal., 2020).
According to the analysis performed by the researchers of the Smart DevOps
project, each job profile requires a different curriculum stemming from four learn-
ing objectives: (a) Development the transversal skills, (b) Building an adequate IT
knowledge background, (c) Developing advanced software development and opera-
tion skills, and (d) Developing smart city management skills (Iatrellis, Panagiotako-
poulos, etal., 2020). The case study focused on the decision node prior to starting
Fig. 9 Smart DevOps Learning pathways
8 https:// smart devops. eu/.

J. Comput. Educ.
1 3
the specialization training in order to evaluate the output of the EDUC8EU subsys-
tem. Initially, the academic advising guidelines for each learning pathway were rep-
resented as a decision tree, since it constitutes a simple way to understand and inter-
pret (Iatrellis, Savvas, etal., 2020). Figure10 depicts the decision tree for the “Smart
city IT manager” learning pathway according to which students were checked by
their personality code, their knowledge and skills, and interest degree. It has to be
noted that each leaf node of a decision tree features a recommendation ID that asso-
ciates the output with an explanatory text documenting the reasoning process. Sub-
sequently, the decision tree was translated into SWRL language and imported into
the EDUC8EU semantic model. The three decision trees were developed by ana-
lyzing the data from popular and heavily researched career inventories such as the
ESCO9 classification, O*Net,10 and ILO.11 For example, the Holland code node in
Fig.10 was defined to “Enterprise” in alignment with the “Information Technology
Project Managers” profile of the O*net library where it has a 96 out of 100 impor-
tance rating. In the O*NET library, the personality type is called “Interests,” but
Fig. 10 Smart city IT Manager academic advising guidelines
9 https:// esco. ec. europa. eu/.
10 https:// www. oneto nline. org/.
11 https:// www. ilo. org/.

1 3
J. Comput. Educ.
it uses the same control measures from Holland’s occupational themes (Realistic,
Investigative, Artistic, Social, Enterprising, Conventional).
For each student, the knowledge that is modeled as part of the semantic model
combined with the student’s parameters triggers the rule execution engine. The
result of the specific interoperation is a recommendation of the learning pathway to
be proposed. In this case study, we used 124 student cases to create the dataset for
validation. More specifically, five academic advisors were engaged in the procedure
with the aim of providing recommendations for finding the learning pathway that
best fits each student. All participants were tenured faculty members with consid-
erable experience in MOOCs and deep knowledge of learning pathways’ require-
ments, prerequisites, and policies and whose experience ranged from 13 to 31years
in Higher Education. The consistency of performance is evaluated using Cohen’s
kappa statistic, which measures the level of agreement between two raters (EDU-
C8EU and academic advisors) by using the term of chance agreement. Cohen’s
kappa coefficient can be calculated as follows:
where P(a) refers to the observed probability of agreement while P(e) is the expected
probability of agreement.
During the execution of the case study, the EDUC8EU system only achieved
fundamental agreement (κ = 0.6231) with the academic advisors, thus a series of
changes and adaptations to the academic advising guidelines incorporated in the
three learning pathways of the MOOC was required. The main cause of the disa-
greement that was identified was that the academic advisors unanimously consid-
ered that the Holland Code assessment decision node should not be used as a root
node, since the lack of IT skills, knowledge, or interest poses a major obstacle to
fulfilling the requirements of under study learning pathways. The academic advi-
sors recognized that this is a strength of the way the EDUC8EU system is conceived
and works since it mimics the human expertise instead of relying solely on easy
and sometimes simplistic personality-based matching techniques that automatically
link to a range of options. Multicriteria matching of recommendations to learner’s
parameters and their detailed and comprehensive analysis that is displayed as the
final output to the student increase the likelihood that the provided advice by EDU-
C8EU will be followed. Thus, with the system’s help, the goal of optimizing the
quality of the academic advising services provided in conjunction with alleviating
any inconsistencies will be achieved. Therefore, using the EDUC8EU backend, the
corresponding changes were made in the decision trees, which for the case of the
“Smart City IT Manger” learning pathway took the following form (see Fig.11):
Once the semantic model was corrected, the evaluation resulted in a higher
concordance with the opinion of the experienced academic advisors. Table5 lists
the results of agreement of the tested cases and the process of calculating Cohen’s
Kappa coefficient is as follows:
Based on Table5, the value of kappa value was κ = 0.8831. The result shows that
EDUC8EU has a strong correspondence with the opinion of the academic advisors.
(4)
𝜅
=
P(a)−P(e)
1
−
P
(
e
)

J. Comput. Educ.
1 3
Furthermore, the usefulness of the overall implemented EDUC8EU software
environment was evaluated by the five academic advisors who were granted full
privileges for the backend system. To mitigate the risk of academic advisors not
understanding the way that backend features should be used, academic advisors
received a 3-h long instruction that was aimed to explain each feature including the
reasoning process, the semantic model, fuzzy logic controller, and the rule genera-
tor. The following seven questions were asked to the academic advisors using a feed-
back questionnaire and 5-level Likert scale where a score of “1” represented strongly
unfavorable to the proposed software solution and a score of "5" represented strongly
Fig.11 Smart city IT Manager modified academic advising guidelines
Table 5 Agreement table Academic advisor
positive
Academic
advisor nega-
tive
EDUC8EU positive 94 3
EDUC8EU negative 2 25

1 3
J. Comput. Educ.
favorable (Table6). We provide the average score that the EDUC8EU received for
each of these questions along with the question:
The participants in the case study felt that the EDUC8EU has a positive effect
on the academic advising process, while at the same time addressing the fuzziness
in learner’s parameters, required for leveraging personalized learning. Moreover, it
was pointed out that the simplicity and effectiveness of the implemented software
environment makes it usable and practical for various educational programs and
scenarios.
Our contribution
Our proposed approach led to the development of EDUC8EU platform, which sup-
ports the provision of highly personalized recommendations for every individual
student by combining fuzzy logic with an evolving expert system and semantic web
technologies. Our platform constitutes a novel combination of fuzzy, ontological,
and rule-based reasoning techniques in academic advising systems that tackle the
following research problems:
(1) The probabilistic nature of student data / outcomes and the vagueness in the
formulation of academic advising recommendations can potentially create uncer-
tainty in an academic decision support architecture (Mohamed Baloul & Wil-
liams, 2013).
(2) The formalization of academic advising recommendations constitutes one of the
major challenges for the area (Shatnawi etal., 2014), (Iatrellis etal., 2017).
(3) From HEI’s perspective, the academic advising recommendations reflect on the
academic, the techno-economic, and the administrative aspects of the higher
education processes (Iatrellis etal., 2018); therefore, it is important to be aca-
demically adaptable to the modern educational trends and responsive to admin-
istrative business process changes or transformations.
Table 6 Evaluation questionnaire
# Question Score
1 It is simple to use the EDUC8EU system 3.8
2 EDUC8EU can effectively support the academic advising process 4.0
3 The method is able to efficiently deal with the uncertainty and fuzziness of the academic
advising process
4.2
4 It was easy to learn to use EDUC8EU backend tools 3.4
5 I believe EDUC8EU can be parameterized to serve various educational programs and sce-
narios
4.8
6 I believe EDUC8E increases the chances of proposing the most appropriate academic advis-
ing guidelines for each student
4.6
7 Overall, I am satisfied with EDUC8EU system 4.4

J. Comput. Educ.
1 3
Our research addresses the first problem by determining a suitable learner model for
representing the important factors affecting decision-making process, through exten-
sive study of the literature and bibliography in the specific domain of our interest.
EDUC8EU approach utilizes fuzzy logic to absorb the vagueness that may exist in
these factors alongside with the imprecision in judgments regarding the student’s
next academic step. The adaptation of academic advising guidelines is performed
by the establishment of a semantic model the core of which is the EDUC8EU ontol-
ogy. EDUC8EU ontology is further utilized for the implementation of a rule-set of
semantic rules. The reasoning process is performed by utilizing an expert system as
the backbone of the EDUC8EU system. Thus, the academic advising workflow for
each student is totally personalized and based on their personality, interests, prior
skills, and knowledge and requirements of the learning pathway offered to them.
The second problem is counter-measured by achieving academic advising guide-
lines formalization, both in terms of content and structure, by means of the semantic
rules, which can facilitate the establishment of a mechanism for the consistent rec-
ommendation of the next appropriate step of an academic plan. The resulting formal
rules can be used as a valuable online enchiridion by the HEI to provide consistent
academic orientation and guidance.
EDUC8EU tackles the third challenge by offering a complete toolkit to the aca-
demic personnel in order to maintain and update the stored rules, thus covering
the complete lifecycle of the academic advising process (both design and execu-
tion mode). In order to facilitate the tuning of the fuzzy parameters in an integrated
way by the domain knowledge experts (academic advisors), an online membership
function editor tool was implemented for creating and editing the membership func-
tions (MFs) for every input and output variables. Moreover, EDUC8EU infrastruc-
ture encompasses an extensible semantic model utilized for the representation of the
required domains of knowledge, as well as for the creation of a set of semantic rules
for the modeling of the academic advising experience and knowledge. The rules are
used in the inference process to discover new facts from given ones and redefine the
rule base of the EDUC8EU platform accordingly.
Conclusions andfuture work
The EDUC8EU software platform offers a “tight integration” of ontological, fuzzy
reasoning and rule-based tools all together. It is implemented exploiting state-of-the-
art technologies, starting from a conventional rule-based expert system and building
upon it to perform fuzzy reasoning. Successful determination of student interest in a
specific learning pathway through fuzzy logic enables the propagation of fuzzy truth
values along the triggered semantic rules. In this way, fuzzy reasoning, conventional
expert system, and semantic web technologies are combined to form a unified rea-
soning process to produce personalized recommendations for the next academic step
in an integrated way. The cornerstone of the EDUC8EU infrastructure is a multi-
facet ontology coupled with the implemented rule-set and the enhanced expressivity
inherited from the SWRL language, which can be easily adapted to serve various
educational needs and scenarios. The development of the expert system was realized

1 3
J. Comput. Educ.
by utilizing the Drools system, while the online membership editor has been devel-
oped as an add-on for the online software platform, a technical decision that greatly
enhances the accuracy and robustness of the system. At the same time, the Drools
rule engine can enrich the semantic model by inducing new implicit knowledge as
each pathway progresses. As rules are fired, new facts are inserted into the fact base,
which can be used in further inference and ontology evolution. Currently, we are
exploiting our system for assisting students on their academic paths in the context
of the INVEST4EXCELLENCE H2020 project that encompasses programs of study
with a wide variety of educational options and experiences in diverse settings.
In future, a type-2 fuzzy approach may be explored to determine the rules and
values of the parameters. Moreover, future work will be devoted to investigate the
adoption of learning analytics techniques to support the decision-making process
through predictive analysis. To this regard, we plan to develop a data-driven predic-
tion model that will utilize machine learning methods to perform clustering accord-
ing to their students’ past decisions on learning pathways.
Acknowledgements The authors would like to thank the INVEST4EXCELLENCE project under the
H2020-IBA-SwafS-Support-2-2020 program (Project No. 101035815, www. invest- allia nce. eu) for pro-
viding support and thank the other project partners.
References
Abdel-Hafez, A., Tang, X., Tian, N., & Xu, Y. (2014). A reputation-enhanced recommender system. In
X. Luo, J. X. Yu, & Z. Li (Eds.), Advanced data mining and applications (pp. 185–198). Springer
International Publishing
Abduldaim, A. M., & Sabri, R. I. (2019). The effectiveness of LUD on digital image watermarking based
on sugeno fuzzy inference system. International Journal of Latest Engineering and Management
Research (IJLEMR), 4, 53–60.
Adomavicius, G., & Tuzhilin, A. (2015). Context-aware recommender systems. In F. Ricci, L. Rokach, &
B. Shapira (Eds.), Recommender systems handbook (2nd ed., pp. 191–226). Springer
Aguilar, J., Valdiviezo-Díaz, P., & Riofrio, G. (2017). A general framework for intelligent recommender
systems. Applied Computing and Informatics, 13, 147–160. https:// doi. org/ 10. 1016/j. aci. 2016. 08.
002
Aly WM, Eskaf KA, Selim AS (2017) Fuzzy mobile expert system for academic advising. In: Canadian
Conference on Electrical and Computer Engineering. pp. 1187–1191
Anderman, E. M., Gray, D. L., & Chang, Y. (2012). Motivation and classroom learning. In I. Weiner
(Ed.), Handbook of psychology (2nd ed.). American Cancer Society
Mohamed Baloul, Williams, P., (2013), Fuzzy academic advising system for on probation students in
colleges of applied sciences. In: International conference on computing, electrical and electronic
engineering (ICCEEE). pp. 372–377
Bielikovà, M., Šimko, M., Barla, M., etal. (2014). ALEF: From application to platform for adaptive col-
laborative learning. Recommender systems for technology enhanced learning: research trends and
applications (pp. 195–225). Springer
Carchiolo, V., Longheu, A., & Malgeri, M. (2010). Reliable peers and useful resources: Searching for the
best personalised learning path in a trust- and recommendation-aware environment. Information Sci-
ences, 180, 1893–1907. https:// doi. org/ 10. 1016/j. ins. 2009. 12. 023
Casali A, Gerling V, Deco C, Bender C (2011) A recommender system for learning objects personalized
retrieval. In: Educational Recommender Systems and Technologies: Practices and Challenges. IGI
Global, pp. 182–210

J. Comput. Educ.
1 3
Chen Y, Pan C, Yang G, Bai J (2014) Intelligent decision system for accessing academic performance of
candidates for early admission to university. In: 10th International Conference on Natural Computa-
tion (ICNC). pp. 687–692
Chen, C.-M., & Duh, L.-J. (2008). Personalized web-based tutoring system based on fuzzy item response
theory. Expert Systems with Applications, 34, 2298–2315. https:// doi. org/ 10. 1016/j. eswa. 2007. 03.
010
Chen, C.-M., Lee, H.-M., & Chen, Y.-H. (2005). Personalized e-learning system using item response
Theory. Computers & Education, 44, 237–255. https:// doi. org/ 10. 1016/j. compe du. 2004. 01. 006
Dias, A. D. S., & Wives, L. K. (2019). Recommender system for learning objects based in the fusion of
social signals, interests, and preferences of learner users in ubiquitous e-learning systems. Personal
and Ubiquitous Computing, 23, 249–268. https:// doi. org/ 10. 1007/ s00779- 018- 01197-7
Díaz-Díaz, J. M., & Galpin, I. (2020). Evaluating models for a higher education course recommender
system using state exam results. Springer
Drachsler, H., Verbert, K., Santos, O. C., & Manouselis, N. (2015). Panorama of recommender systems to
support learning. In F. Ricci, L. Rokach, & B. Shapira (Eds.), Recommender systems handbook (pp.
421–451). Springer
du Boulay, B., Avramides, K., Luckin, R., Martínez-Mirón, E., Méndez, G. R., & Carr, A. (2010).
Towards systems that care: A conceptual framework based on motivation, metacognition and affect.
International Journal of Artificial Intelligence in Education, 20, 197–229. https:// doi. org/ 10. 3233/
JAI- 2010- 0007
Duarte, R., de Oliveira Pires, A. L., & Nobre, Â. L. (2018). Mature learners’ participation in higher edu-
cation and flexible learning pathways: Lessons learned from an exploratory experimental research.
In M. M. Nascimento, G. R. Alves, & E. V. A. Morais (Eds.), Contributions to higher engineering
education (pp. 33–53). Springer
Durao, F., Dolog, P., (2009). Social and behavioral aspects of a tag-based recommender system. In: ISDA
2009—9th International Conference on Intelligent Systems Design and Applications. pp. 294–299
Eccles, J. S., (1983). Expectancies, values, and academic behavior. Achievement and achievement
motives: Psychological and sociological approaches. pp. 75–146
Essa, A. (2016). A possible future for next generation adaptive learning systems. Smart Learning Envi-
ronments, 3, 16. https:// doi. org/ 10. 1186/ s40561- 016- 0038-y
Fallahnejad, M., & Moshiri, B. (2014). The performance of B-spline and gaussian functions in the struc-
ture of a Neuro-Fuzzy network. Technical and Vocational University, 4, 1622–1636.
Farzan, R., & Brusilovsky, P. (2006). Social navigation support in a course recommendation system. In
V. P. Wade, H. Ashman, & B. Smyth (Eds.), Lecture notes in computer science (including subseries
Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (pp. 91–100). Springer
Garrido, A., & Morales, L. (2014). E-Learning and intelligent planning: Improving content personali-
zation. Revista Iberoamericana De Tecnologias Del Aprendizaje, 9, 1–7. https:// doi. org/ 10. 1109/
RITA. 2014. 23018 86
Henderson, L. K., & Goodridge, W. (2015). AdviseMe: An intelligent web-based application for aca-
demic advising. (IJACSA) International Journal of Advanced Computer Science and Applications.
https:// doi. org/ 10. 14569/ IJACSA. 2015. 060831
Horrocks, I., Patel-Schneider, P. F., Boley, H., etal (2010) SWRL: A semantic web rule language com-
bining OWL and RuleML. In: W3C Member Submission. Retrieved January 30, 2017, from https://
www. w3. org/ Submi ssion/ SWRL/
Iatrellis, O., Kameas, A., & Fitsilis, P. (2017). Academic advising systems: A systematic literature review
of empirical evidence. Education Sciences, 7, 90. https:// doi. org/ 10. 3390/ educs ci704 0090
Iatrellis, O., Kameas, A., & Fitsilis, P. (2018). EDUC8: Self-evolving and personalized learning pathways
utilizing semantics. IEEE Conference on Evolving and Adaptive Intelligent Systems (EAIS), 2018,
1–8.
Iatrellis, O., Kameas, A., & Fitsilis, P. (2019a). A novel integrated approach to the execution of personal-
ized and self-evolving learning pathways. Education and Information Technologies. https:// doi. org/
10. 1007/ s10639- 018- 9802-7
Iatrellis, O., Kameas, A., & Fitsilis, P. (2019b). EDUC8 ontology: Semantic modeling of multi-
facet learning pathways. Education and Information Technologies. https:// doi. org/ 10. 1007/
s10639- 019- 09877-4
Iatrellis, O., Panagiotakopoulos, T., Gerogiannis, V. C., etal. (2020). Cloud computing and semantic
web technologies for ubiquitous management of smart cities-related competences. Education and
Information Technologies. https:// doi. org/ 10. 1007/ s10639- 020- 10351-9

1 3
J. Comput. Educ.
Iatrellis, O., Savvas, I. K., Kameas, A., & Fitsilis, P. (2020). Integrated learning pathways in higher
education: A framework enhanced with machine learning and semantics. Education and Infor-
mation Technologies. https:// doi. org/ 10. 1007/ s10639- 020- 10105-7
Imran, H., Belghis-Zadeh, M., Chang, T.-W., Kinshuk, & Graf, S. (2016). PLORS: A personalized
learning object recommender system. Vietnam Journal of Computer Science, 3, 3–13. https:// doi.
org/ 10. 1007/ s40595- 015- 0049-6
Irfan, M., Alam, C. N., & Tresna, D. (2019). Implementation of fuzzy mamdani logic method for
student drop out status analytics. Journal of Physics: Conference Series. https:// doi. org/ 10. 1088/
1742- 6596/ 1363/1/ 012056
Kaklauskas, A., Zavadskas, E. K., Seniut, M., Stankevic, V., Raistenskis, J., Simkevičius, C., Stank-
evic, T., Matuliauskaite, A., Bartkiene, L., Zemeckyte, L., Paliskiene, R., Cerkauskiene, R., &
Gribniak, V. (2013). Recommender system to analyze student’s academic performance. Expert
Systems with Applications, 40, 6150–6165. https:// doi. org/ 10. 1016/j. eswa. 2013. 05. 034
Kaufmann, H. R., Bengoa, D., Sandbrink, C., Kokkinaki, A., Kameas, A., Valentini, A., & Omiros, I.
(2020). DevOps competences for smart city administrators. CORP, 2020, 213–223.
Kerkiri, T., Manitsaris, A., Mavridou, A., (2008). Reputation metadata for recommending personal-
ized e-learning resources. In: Second International Workshop on Semantic Media Adaptation
and Personalization. pp. 110–115
Luyi, Li., Yanlin, Z., Ogata, H., Yano, Y., (2004). A framework of ubiquitous learning environment.
In: The Fourth International Conference on Computer and Information Technology. pp. 345–350
Martín, E., & Carro, R. M. (2009). Supporting the development of mobile adaptive learning environ-
ments: A case study. IEEE Transactions on Learning Technologies, 2, 23–36. https:// doi. org/ 10.
1109/ TLT. 2008. 24
McCarthy, W. E. (2003). The REA modeling approach to teaching accounting information systems.
Issues in Accounting Education, 18, 427–441. https:// doi. org/ 10. 2308/ iace. 2003. 18.4. 427
Medsker, L. R. (1995). Hybrid intelligent systems. Springer
Molina-Solana, M., Birch, D., & Guo, Y. K. (2017). Improving data exploration in graphs with fuzzy
logic and large-scale visualisation. Applied Soft Computing Journal, 53, 227–235. https:// doi.
org/ 10. 1016/j. asoc. 2016. 12. 044
Nauta, M. M. (2010). The development, evolution, and status of Holland’s theory of vocational per-
sonalities: Reflections and future directions for counseling psychology. Journal of Counseling
Psychology, 57, 11–22. https:// doi. org/ 10. 1037/ a0018 213
Park, D. H., Kim, H. K., Choi, I. Y., & Kim, J. K. (2012). A literature review and classification of
recommender systems research. Expert Systems with Applications, 39, 10059–10072. https:// doi.
org/ 10. 1016/j. eswa. 2012. 02. 038
Pintrich, P. (2003). Motivation and classroom learning. In I. B. Weiner (Ed.), Handbook of psychol-
ogy. Wiley
Prasad, M., Liu, Y. T., Li, D. L., etal. (2017). A new mechanism for data visualization with TSK-type
preprocessed collaborative fuzzy rule based system. Journal of Artificial Intelligence and Soft
Computing Research, 7, 33–46. https:// doi. org/ 10. 1515/ jaiscr- 2017- 0003
Ricci, F., Shapira, B., & Rokach, L. (2015). Recommender systems handbook (2nd ed.). Springer
Ryan, R. M., & Deci, E. L. (2000). Intrinsic and extrinsic motivations: Classic definitions and new
directions. Contemporary Educational Psychology, 25, 54–67. https:// doi. org/ 10. 1006/ ceps.
1999. 1020
Salehi, M., & Kmalabadi, I. N. (2012). A hybrid attribute–based recommender system for e–learning
material recommendation. IERI Procedia, 2, 565–570. https:// doi. org/ 10. 1016/j. ieri. 2012. 06. 135
Santos, O. C., Boticario, J. G., & Pérez-Marín, D. (2014). Extending web-based educational systems
with personalised support through user centred designed recommendations along the e-learning
life cycle. Science of Computer Programming, 88, 92–109. https:// doi. org/ 10. 1016/j. scico. 2013.
12. 004
Schoefegger, K., Seitlinger, P., & Ley, T. (2010). Towards a user model for personalized recommenda-
tions in work-integrated learning: A report on an experimental study with a collaborative tagging
system. Procedia Computer Science, 1, 2829–2838.
Shatnawi, R., Althebyan, Q., Ghalib, B., Al-Maolegi, M., (2014). Building a smart academic advising
system using association rule mining
Takano, K., Li, K. F., (2009). An adaptive personalized recommender based on web-browsing behav-
ior learning. In: Proceedings—International Conference on Advanced Information Networking
and Applications, AINA. pp. 654–660

J. Comput. Educ.
1 3
Thanh-Nhan, H-L., Nguyen, H-H., Thai-Nghe, N., (2016). Methods for building course recommenda-
tion systems. In: 2016 Eighth International Conference on Knowledge and Systems Engineering
{KSE}. pp. 163–168
Troussas, C., Krouska, A., & Virvou, M. (2020). Using a Mult module model for learning analytics to
predict learners’ cognitive states and provide tailored learning pathways and assessment. In M.
Virvou, E. Alepis, G. A. Tsihrintzis, & L. C. Jain (Eds.), Machine learning paradigms: Advances
in learning analytics (pp. 9–22). Springer International Publishing.
Upendran, D., Chatterjee, S., Sindhumol, S., & Bijlani, K. (2016). Application of predictive analytics
in intelligent course recommendation. Procedia Computer Science, 93, 917–923. https:// doi. org/
10. 1016/j. procs. 2016. 07. 267
Wigfield, A., & Cambria, J. (2010). Students’ achievement values, goal orientations, and interest: Def-
initions, development, and relations to achievement outcomes. Developmental Review, 30, 1–35.
https:// doi. org/ 10. 1016/j. dr. 2009. 12. 001
Xu, J., Xing, T., & van der Schaar, M. (2016). Personalized course sequence recommendations. IEEE
Transactions on Signal Processing, 64, 5340–5352. https:// doi. org/ 10. 1109/ TSP. 2016. 25954 95
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
Omiros Iatrellis has a Ph.D. in Computer Science from the School of Science & Technology, Hellenic
Open University (Greece) and an M.Sc. in Computer Networks from the University of Middlesex (UK).
He is also a graduate of the Physics department of the Ioannina University. He is a Lecturer at the depart-
ment of Digital Systems of the University of Thessaly, Greece. He publishes regularly in peer reviewed
international journals and conferences in the area of software engineering, semantic web, and education.
Also, he participated in various research projects of University of Thessaly.
Evangelos Stamatiadis holds a B.Sc. in Computer Systems Engineering from Technological Higher
Educational Institution of Piraeus, an M.Sc. in Information and Communication Systems from Open Uni-
versity of Cyprus, and he is a candidate Ph.D. student at the University of Thessaly. He speaks English
and French. Mr. Stamatiadis has over 22years of experience in managing complex ICT environments. He
specializes in cloud-based information systems and enterprise architecture. He has authored books for the
National Centre of Public Administration and Local Government where he has taught for 15years.
Nicholas Samaras (IEEE SM) is a Professor at the Department of Digital Systems, at the University
of Thessaly, in Larissa, Greece. Dr. Samaras received his Ph.D. degree from the University of Pittsburgh,
in Pittsburgh, PA, USA, in electrical engineering. His current research interests include IoT Systems and
Applications, Networked Control Systems, and Industrial Automation. He has served in several organ-
izing, Steering, and/or Program Committees, for several international conferences and he is an Associate
Editor and paper reviewer for various International Journals. He was a co-recipient of the IEEE Industry
Applications Society Prize Paper Award in 1998.
Theodor Panagiotakopoulos received his Diploma and Ph.D. from the Department of Electrical and
Computer Engineering, University of Patras, Greece, in 2006 and 2011, respectively. He is currently a
senior research fellow in the Mobile and Pervasive Computing, Quality and Ambient Intelligence Labora-
tory, School of Science and Technology, Hellenic Open University. His research interests include perva-
sive computing, internet of things, ambient intelligence, smart city applications, ambient-assisted living,
mobile health, fuzzy systems, instructional design and development, e-learning platforms, and digital
literacy. He has published more than 25 scientific articles in international book chapters, journals, and
conferences and has participated in 12 European and National research programs holding key positions at
a research, technical, and managerial level.
Panos Fitsilis is a full Professor at Business Administration Department of the University of Thessaly,
Greec., and academic coordinator of the module “Software Design” at Hellenic Open University. He has
extensive project management experience with the development and deployment of large IT systems and
extensive management experience in various senior management positions. His research interests include

1 3
J. Comput. Educ.
Smart Cities, Smart Factories, Business Information Systems, Social, Educational Technology, Software
Project Management, and so on.