A Technology-enhanced Smart Learning Environment based on the

Combination of Knowledge Graphs and Learning Paths

Eleni Ilkou

and Beat Signer

Web & Information Systems Engineering Lab, Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium

Keywords:

Knowledge Graphs, Learning Paths, e-Learning, Smart Education, Smart Learning, Educational Application,

Assessment Classiﬁcation, Personalised Teaching Assistant Tool, Mathematics Education.

Abstract:

In our position paper on a technology-enhanced smart learning environment, we propose the innovative com-

bination of a knowledge graph representing what one has to learn and a learning path deﬁning in which order

things are going to be learned. In this way, we aim to identify students’ weak spots or knowledge gaps in

order to individually assist them in reaching their goals. Based on the performance of different learning paths,

one might further identify the characteristics of a learning system that leads to successful students. In addi-

tion, by studying assessments and the different ways a particular problem can be solved, new methods for a

multi-dimensional classiﬁcation of assessments can be developed. The theoretical ﬁndings on learning paths in

combination with the classiﬁcation of assessments will inform the design and development of a smart learning

environment. By combining a knowledge graph with different learning paths and the corresponding practical

assessments we enable the creation of a smart learning tool. While the proposed approach can be applied to

different educational domains and should lead to more effective learning environments fostering deep learning

in schools as well as in professional settings, in this paper we focus on the domain of mathematics in primary

and high schools as the main use case.

1 INTRODUCTION

Have you always been good in maths or physics? Did

you have top grades in all your classes or were there

one or two classes that you did not particularity like?

If that is the case, then the proposed approach aims to

provide some insights on why this might happen and

how the problem can be addressed. A major issue in

today’s educational systems is the restricted amount

of individual and customised support that students get

from their teachers. This is based on the problem that

students can often not identify their own knowledge

gaps since they usually do not know what they do not

know, making it difﬁcult for educators to assist them

properly.

An approach for solving the problem is the cre-

ation of a knowledge graph

, a semantic representa-

tion of all the knowledge for a given domain and the

associations (links) between different topics (Stapel

https://orcid.org/0000-0002-4847-6177

https://orcid.org/0000-0001-9916-0837

https://www.blog.google/products/search/introducing-

knowledge-graph-things-not/

et al., 2016; Rizun, 2019; Lecailliez et al., 2019).

By using a knowledge graph, one can identify any

necessary prerequisite knowledge (prerequisite rela-

tions) for a given topic and track the areas where a

student seems to have a lack of performance. It is

obvious that the general knowledge about a domain

(e.g. mathematics) is linked in a unique way with

a single representation (knowledge graph). Any ad-

ditional metadata and information that is necessary

to support the learning process might refer to such

a knowledge graph or be implemented to top of the

knowledge graph.

However, a knowledge graph on its own is not

enough to point out a student’s potential knowledge

gaps. Often students are changing learning environ-

ments or moving from one school with a certain ed-

ucational policy to another school with a different

policy. In such a situation, students are facing a

new educational system, which makes it difﬁcult for

them to perform decently and often leads to school

dropouts (South et al., 2007). These problems are

based on the fact that each school follows a speciﬁc

curriculum, which can differ signiﬁcantly from other

curricula. In order to investigate and study the dif-

Ilkou, E. and Signer, B.

A Technology-enhanced Smart Learning Environment based on the Combination of Knowledge Graphs and Learning Paths.

DOI: 10.5220/0009575104610468

In Proceedings of the 12th International Conference on Computer Supported Education (CSEDU 2020) - Volume 2, pages 461-468

ISBN: 978-989-758-417-6

461

Learning Path

Square root of

positive integer

numbers

ex.

49 7

Square power of

positive integer

numbers

ex.

7 49

Multiplication of

positive integer

numbers

ex.

7 7 49

× =

l k

b b b

× × ×

…

contains contains

contains

Knowledge Graph

prerequisite link

links to

Figure 1: Proposed combination of a learning path and a knowledge graph.

ferent systems, the concept of a learning path

has

been introduced. The learning path of each school

represents the order in which the knowledge is be-

ing taught—or in other words—how the curriculum

traverses the knowledge graph. The learning paths a

student has followed in combination with the knowl-

edge graph, as illustrated in Figure 1, could poten-

tially shed light on a student’s secret of success.

Since there are different learning paths result-

ing from different curricula and educational policies,

there should also be a variation in the success of these

learning paths. A question we might ask ourselves is

whether there are good and bad educational policies.

If some learning paths are leading to more successful

students, the question is whether we could identify the

characteristics of a successful learning path.

Surprisingly, we often see some confusion be-

tween the concept of a knowledge graph and a learn-

ing path. KnowEdu (Chen et al., 2018b) is an au-

tomatic knowledge graph generation tool for educa-

tional purposes, which extracts the concepts of sub-

jects or courses from textbooks and exams, and iden-

tiﬁes the educational relations between the concepts.

Similar to KnowEdu, the K12EduKG system (Chen

et al., 2018a) uses the same technique to construct

a “knowledge graph”, but with the objective of aid-

ing and improving the ﬂow of teaching and learning,

rather than deﬁning course dependencies. The com-

mercial Mathspace

solution is an online platform

for mathematics making use of a knowledge graph

for topics of the U.S. math curriculum and providing

teaching assistance for students as well as teachers.

As mentioned above, the problem with some of

the presented existing approaches is that even in the

scientiﬁc and business world, there seems to be some

confusion between the concepts of a knowledge graph

and a learning path. For example, all the solutions

https://en.wikipedia.org/wiki/Learning pathway

https://mathspace.co

mentioned in the previous paragraph created learning

paths based on the speciﬁc curriculum of the available

teaching material, and not a knowledge graph as they

state.

Besides a series of theoretical concepts, a part of

the learning path is formed by practical assessments,

which differ between learning paths in terms of their

content and the speciﬁc solving approach. Often stu-

dents are asked to approach a knowledge topic in mul-

tiple different ways in order to be able to apply the

theory and multiple methodologies per topic. These

methodologies can address different types of assess-

ments and solving approaches, such as assessments

solved by calculation or assessments solved by proof.

The use of different solving approaches might deter-

mine the characteristics of a successful learning path.

It is our interest to create a model that can map these

correlations between methodologies and groups of as-

sessments. Our approach is trying to integrate the

learning outcome with the knowledge topic and to fo-

cus on the methodologies that are being taught for the

different curricula and their learning paths, in order to

construct a knowledge representation that covers all

the knowledge components that are applied during the

teaching process. Moreover, solving a problem with

multiple methods can be beneﬁcial for the knowledge

development of students (Cai and Nie, 2007) as well

as teachers (Levav-Waynberg and Leikin, 2006).

While there is plenty of educational material avail-

able online, the content is often unstructured and not

well connected. Practical assessments for the differ-

ent knowledge domains are usually introduced to stu-

dents during their study of a speciﬁc topic. How-

ever, complex assessments with a high difﬁculty level

might require knowledge from multiple domains in

order to be solved. This can mislead students and

make them believe that they do not have the knowl-

edge about a given topic—such as how to compute

the area of a rectangle as illustrated in Figure 5—

CSEDU 2020 - 12th International Conference on Computer Supported Education

462

although in reality their knowledge gap is in another

required knowledge domain (e.g. quadratic equations

in the example).

2 SMART LEARNING

ENVIRONMENT

We propose the construction of a smart learning en-

vironment that can address the problems presented

in the previous section. The system will contain

and link educational material for a given domain,

such as chemistry, mathematics or history. The sys-

tem is designed around one or multiple knowledge

graphs, which can be connected; a physics knowledge

graph for instance shares concepts with a mathemat-

ics knowledge graph hence there are interconnected

topics. On top of the knowledge graphs, we represent

the curricula (learning paths) of different schools the

students are following. The learning paths can help us

to provide personalised aid to students. Besides deﬁn-

ing the dependencies between knowledge topics in a

knowledge graph, individual assessments should be

linked to the corresponding knowledge topics. There-

fore there is an interest in developing an assessment

classiﬁcation based on the different characteristics of

assessments. Eventually, the results of our research

should become available in the form of a smart learn-

ing tool supporting learners as well as teachers, as il-

lustrated in Figure 6.

In this position paper, we discuss the potential

next steps in research and development for educa-

tional material as described in the book “How People

Learn” (Council, 2000). We focus on the examina-

tion of common practice by reviewing existing curric-

ula and assessments, the extension of the knowledge

base by proposing the development of new educa-

tional material, and the development of smart learning

tools based on key research ﬁndings and the princi-

ples of learning. We further investigate successful and

creative educational practices and explore the foun-

dations of learning to support new research on assess-

ments that focus on improving teaching and achieving

a deeper understanding of a speciﬁc domain.

2.1 Knowledge Graphs

Educational data has common characteristics as it

usually forms hierarchical structures going from the

simplest knowledge topic to the more advanced ones,

and from the earliest event to the most recent. With

that in our mind, we can construct a knowledge

representation where knowledge topics are arranged

in a directed graph. The graph consists of nodes

representing the knowledge topics and their com-

ponents, as well as links representing the relation-

ships between speciﬁc knowledge topics. A frame-

work for the construction of such a graph based on

the resource-selector-link (RSL) hypermedia meta-

model (Signer and Norrie, 2007) has been proposed

in EduKnow (Ilkou, 2019).

The knowledge graph aims to reveal the inner

structure of the knowledge topics, by handling the

core of the educational material and teaching process,

which is represented by the learning outcomes. By fo-

cussing on the learners and their deep understanding

of a domain, we shall deﬁne rules of how the knowl-

edge topics are related to each other and ﬁnally cre-

ate an abstract model of a graph representation for an

individual educational domain. For that purpose, we

need to have a well-deﬁned structure that can be used

in the same way for each learner and at the same time

be enriched with the necessary elements to represent

all knowledge components. The model needs to be

deﬁned theoretically in order to have an abstract foun-

dation for future research. Furthermore, a theoretical

foundation of our knowledge components and their

relationships is necessary if we thrive to achieve re-

producible data and a framework which can be used

for knowledge representation in arbitrary educational

domains.

Our aim is to develop the foundations for an adap-

tive learning recommendation system. The core of

such a system forms a smart educational application

containing a knowledge graph representation for the

educational domains. Based on this knowledge repre-

sentation it becomes possible to implement the learn-

ing paths of different curricula and to also track a

learner’s knowledge gaps. Aleks

which is based on

the book entitled Knowledge Spaces (Doignon and

Falmagne, 1999), uses the concept of learning paths

to guide the students based on the curriculum they are

following. It is unclear how exactly the knowledge

graph space is created, but we know that the nodes

represent subjects from the U.S. curriculum. The dif-

ference between the proposed adaptive learning rec-

ommendation system and existing solutions is that

adaptive learning will create a knowledge representa-

tion based on how the knowledge itself is structured,

and not by some speciﬁc curricula. Hence, we will

use the knowledge graph “forwards” (in the direction

of the directed links) to ﬁnd the next proposed topics

for reaching a knowledge goal, but also “backwards”

(opposite to the direction of the directed links) to dis-

cover a student’s knowledge gaps. By taking into con-

sideration the mistakes that a learner is making, we

can probabilistically determine the prerequisite topics

https://www.aleks.com

A Technology-enhanced Smart Learning Environment based on the Combination of Knowledge Graphs and Learning Paths

463

that might form the source of a learner’s poor perfor-

mance. Moreover, we aim to visualise the knowledge

structure in order that the actors involved in the learn-

ing process are able to view and understand the rea-

sons behind a student’s performance and the possible

way to success. Note that this novelty in knowledge

representations and knowledge structures represents

one of our main contributions.

Let us have a look at an example for the domain of

algebra taken from the 8th grade mathematics corpus

of the Greek curriculum (B’ Gymnasiou). This ex-

ample brings light to why it is important to separate

the knowledge graph from the learning paths. Fig-

ure 2 shows parts of the curriculum (learning path)

for the knowledge of ‘linear functions’ and the cor-

responding knowledge topics of the knowledge graph

are illustrated in Figure 3. We can see that the learn-

ing path, which is a sequence of knowledge topics

does not correspond to how the knowledge is struc-

tured and does not reveal all the knowledge connec-

tions necessary to reach the end goal. This knowl-

edge graph can help in detecting knowledge gaps by

querying the links for a speciﬁc knowledge topic as

discussed in EduKnow (Ilkou, 2019) and shown in

Figure 3 for the topic of ‘linear functions’. In our ex-

ample, the knowledge topics that correspond to Fig-

ure 2 are represented in blue. There are further differ-

ent types of links representing different kinds of re-

lationships between knowledge topics. We can create

a subgraph of parts of the database, such as the one

shown in Figure 3 with the ‘linear functions’ query

object as a ﬁnal node. This visualisation might as-

sist the student and their teacher to ﬁnd all the possi-

ble reasons (non-mastered topics) for a student’s poor

understanding of a given topic.

Table of contents

1. Equations – Inequalities

1.1. Algebraic expressions

1.2. 1

degree equations

1.3. Solving formulas

1.4. Solving problems with equations

1.5. 1

degree inequalities

2. Real Numbers

2.1. Square root of a positive number

2.2. Irrational and real numbers

2.3. Problems

3. Functions

3.1. Definition of a function

3.2. Graph of a function

3.3. Function y=ax

3.4. Function y=ax+b

Figure 2: Part of the table of contents for the 8th grade

mathematics corpus of the Greek curriculum showing the

learning path leading to the knowledge of ‘linear functions’.

Equality

and

Inequalities

Linear

functions

1st degree

inequalities

1st degree

equations

Powers

and

Squares

Real

Numbers

Rational

Numbers

Alebraic

Expres-

sions

Fractions

Whole

Numbers

Decimal

Numbers

Natural

Numbers

Function

Definition

and Graph

Function

Domain

and Image

Analogy

Figure 3: Parts of the knowledge graph for the 8th grade

mathematics corpus of the Greek curriculum highlighting

the knowledge components for the topic of ‘linear func-

tions’.

We treat a knowledge graph as a multidimensional

representation which connects knowledge topics but

also contains all the necessary knowledge compo-

nents for a given knowledge topic. As mentioned pre-

viously, a knowledge topic is a node in the knowledge

graph and consists of the title of the topic as the high-

est level of abstraction, and the methodologies, the-

ory and solved examples at a more detailed level. In

the suggested model, knowledge is the outcome of the

learning process and represents what a student should

know by the end of a chapter or academic year. There

is often confusion for existing knowledge represen-

tation techniques on what should be represented as a

node in the knowledge graph, which is another dif-

ference between our proposal and existing knowledge

mapping techniques. In existing solutions, the nodes

are many times formed by knowledge topics with very

generic content, such as numbers (Chen et al., 2018b),

or with very speciﬁc content such as the chapters

of a curriculum (e.g. Consumer Arithmetic (11-12)

Therefore, it is of great importance to already have

properly identiﬁed all the knowledge topics before

beginning with the linking process. Hence, this task

is carried out by a domain expert who can formulate

each knowledge topic based on speciﬁc learning out-

comes.

Each knowledge topic should have a title and at

least one methodology, which explains how an as-

sessment can be addressed in an abstract way together

with the illustration of a solved example. The method-

ology is many times found as part of the introduction

of a new chapter for a given subject and forms an im-

portant component. By analysing different curricula

https://mathspace.co

CSEDU 2020 - 12th International Conference on Computer Supported Education

464

2nd degree equations

Vieta’s

Formula

Theory

When a is not 0, there are two

solutions to

and they are

is the discriminator

b ac

−

ax bx c

+ + =

b b ac

− ± −

Theory

When with the

solutions we know that

ax bx c

+ + =

1 2

x x

1 2

x x

−

+ =

1 2

x x

⋅ =

Theory

When

it can be written as

ax bx c

+ +

( )

ax h k− +

Discriminator

p,q

Methodology

Completing

the Square

Theory

When the

solutions are

x px q

+ + =

2 4

p p

x q

−

= ± −

Figure 4: Visualisation of the knowledge components for the knowledge topic of ‘2nd degree equations’.

in the domain of mathematics, we found that students

are often being taught the same topics, such as multi-

plication or solving quadratic equations, but they are

introduced to a speciﬁc technique (methodology) to

address assessments which might vary signiﬁcantly

for different curricula. Moreover, based on the focus

of the curriculum, students are asked to perform cer-

tain tasks for a given topic, such as doing calculations

to ﬁnd a variable, or to learn only the theory of it. In

other cases, students are asked to approach a knowl-

edge topic from multiple ways in order to be able

to apply the theory and multiple methodologies for a

given topic. These methodologies can address differ-

ent types of assessments, such as assessments solved

by calculations and assessments solved by proof. It is

in our interest to create a model that can map these

correlations between methodologies and groups of

assessments. Our approach is trying to include the

learning outcomes of a knowledge topic and to fo-

cus on the methodologies that are being taught for the

different curricula and learning paths to construct a

knowledge representation that contains all the knowl-

edge components that are used during the teaching

process. An example for the domain of mathematics

showing all the knowledge components of the knowl-

edge topics of ‘2nd degree equations’ is shown in Fig-

ure 4. In this example, a knowledge topic consists of

different methodologies. Each methodology is repre-

sented as a branch of the root 2nd degree equations

and contains the methodology name, the theory and

the solved example (not shown in the ﬁgure).

In contrast to existing techniques, our approach

represents the knowledge nodes as the learning out-

comes, due to the methodologies describing the ap-

proaches to address the assessments. Our novel ap-

proach handles educational data from the learner’s

point of view and lets them know what exactly they

should be able to do in order to succeed. We ﬁnd

this format necessary to achieve high scalability and a

reproduction of the same knowledge structure. Exist-

ing models do not always result in the same knowl-

edge representation, as they are strongly related to

the educational material they are processing. Fur-

ther, these techniques demand a great number of ed-

ucational data in order to create a decent graph. In

the proposed construction of the knowledge graph, all

we need is an expert in the ﬁeld, such as a teacher,

in order to ﬁrst properly identify the knowledge top-

ics, and then associate them via the corresponding

links. By the time this procedure is completed, the

resulting knowledge representation can be reused and

enriched by others. Other experts can build on the

existing knowledge graph by adding new methodolo-

gies or missing knowledge topics and links. The rea-

son for the possible reproduction is the fundamental

structure of the knowledge graph. Since the graph

is structuring pure knowledge without taking into ac-

count any learning paths (e.g. from textbooks or cur-

ricula), it represents fundamental knowledge that will

not change from one institution to another. The only

thing that might happen is that a curriculum addresses

assessments with a different technique, or offers ex-

tended courses that cover extra material that is not yet

implemented by the knowledge representation of the

knowledge graph.

On the practical side, another difference to the ex-

isting body of work is the association of assessments

with the corresponding knowledge topics. We link as-

sessments with the knowledge domains they depend

on in order to be fully addressed. These connections

to the necessary knowledge domains are not always

obvious as illustrated in the example shown in Fig-

ure 5. In order to ﬁnd the right answer in this exam-

ple, a student needs to have knowledge about 2nd de-

gree equations, the area of a rectangle and some ba-

sic principles of geometry. This assessment illustrates

A Technology-enhanced Smart Learning Environment based on the Combination of Knowledge Graphs and Learning Paths

465

the importance of linking an assessment with all its re-

quired knowledge topics and also reveals that the con-

nections of the knowledge topics should be based on

theoretical as well as practical (assessments) aspects.

−

Question: Find x if the area is

Solution:

Area wl

[area of rectangle]

(

)

3 2

x x

= −

[variables]

3 2

x x

= −

3 1 2 1

x x

+ = − +

[2nd degree equations]

( )

4 1

= −

3 1

x or x

= = −

2 1 2 1

x or x

= − − = −

[principles of geometry]

so,

Figure 5: Complex assessment requiring knowledge from

multiple domains.

Assessments have been studied for performance-

based applications (Linn et al., 1991), been classi-

ﬁed in existing intelligent learning environments (Le

and Pinkwart, 2014), and there exist automatic sys-

tems for the recommendation of assessments (Har-

jula, 2008). However, complex assessments that re-

quire knowledge from multiple domains to be solved

are not well linked to those domains. Usually, an

assessment belongs to a single knowledge topic and

therefore only inherits any prerequisite connections

that the corresponding knowledge topic has. This is

also the case that many students are facing, as they

solve simple assessments but cannot perform more

advanced ones since the advanced assessments re-

quire knowledge from multiple domains. In this case,

these connections are not obvious to students and

hence they cannot realise that an assessment needs

extra knowledge (and which extra knowledge) to be

solved. Therefore, in the case of a complex as-

sessment, a student might not perform well without

knowing the reason for their bad performance. With

the proposed knowledge graph approach, a learner

is aware of the complexity of an assessment and all

the required knowledge. An important aspect of the

proposed approach is therefore that it reveals the hid-

den connections of knowledge that exist on a practical

level.

2.2 Encoding of Learning Paths

Another important part of the proposed approach is

the study of learning paths which will be encoded

on top of the knowledge graphs. There is an interest

in researching the different learning paths since often

students with different learning paths are asked to take

the same ﬁnal exams, which might have a major im-

pact on their future career (baccalaureate diploma and

multidisciplinary courses). Our study should provide

more insights into the different used curricula, iden-

tify their strong and loose connections between top-

ics, and help students in higher education who wish to

follow courses from different departments. The goal

is to identify these learning paths and to encode their

content (components), compare them, and help stu-

dents understand how their problem matches course

prerequisites, and keep track of their performance and

knowledge. Students would further understand the

goal of each course. Seeing the big picture will help

them to increase their performance and prevent them

from dropping out, as well as help them to faster adapt

to the courses they are following.

The proposed research is challenging, given that

it requires a formulation of the main components of

each learning path, and inventive when it comes to the

study of the strong and successful as well as the weak

points of curricula following different learning paths

that ultimately lead to the same diploma. We foresee

to contribute to existing knowledge with a number of

innovative aspects for educational systems and their

performance.

The results of our research might assist future ed-

ucation counsellors to make decisions based on the

performance of different learning paths when design-

ing a study path, or to extend the curriculum for any

sector that is related to teaching or learning. Our study

will also help companies to formulate an expertise

learning path based on the knowledge their employ-

ees should possess in order to increase their produc-

tivity (

Zeljko

Sundov and Gregori

c, 2014).

2.3 Assessments Classiﬁcation

After having created the learning paths, we will need

to enrich them with assessments. We propose two

parts for the assessment analysis, based on solving

approaches and the knowledge domains an exercise is

referring to. Regarding the ﬁrst case, we are planning

to create metadata about different solving ways which

will be an innovative feature to be used in an assess-

ment classiﬁer. Moreover, an assessment classiﬁer

tool could classify assessments based on the prerequi-

site knowledge that is needed to solve them. Based on

our idea of combining the learning path with a knowl-

edge graph, a new classiﬁer can follow the link to

the knowledge graph, get the prerequisites for a given

topic and thereby identify all prerequisite knowledge

for a given assessment. An example is the exercise in

which the square power of 7 is 7

= 49 has a prerequi-

site set of exercises (7 × 7 = 49). There will be some

challenges when classifying complex exercises as we

need a multi-dimensional classiﬁer. Hence, the re-

search on the solving ways and data retrieved from the

CSEDU 2020 - 12th International Conference on Computer Supported Education

466

associated knowledge graph will enable us to come

up with a new assessment classiﬁcation model. By

classifying the provided material in different clusters,

we will be able to study the relations between topics

and identify optimal routes for learning paths, as there

will be associations between assessment domains that

do not exist as a clear sequence in the learning path,

such as in Figure 5 between quadratic equations and

the calculation of an area. Also, on the student side,

we will be able to provide a larger variety and quan-

tity of assessments for a given topic to be studied.

The results of such a study will beneﬁt students who

are preparing for their ﬁnal exams, teachers, as well

as companies working in the domain of educational

technologies since they will be able to automatically

classify and share their advanced learning material.

2.4 Smart Learning Tool

A smart learning tool could enable the centralisation

of online and digital exercises to exist in a common

pool with more people being able to beneﬁt from. It

might also help students to identify in a fast and ac-

curate way their knowledge gaps and the precise goal

each student needs to reach in order to succeed in each

grade, based on a school’s curriculum, and track their

performance based on their knowledge and not only

their grades.

All the previous theoretical results will form the

basis for the development of a smart learning tool that

can potentially help students and teachers in real time

and consists of the components shown in Figure 6.

A teacher or tutor is often lacking time for a full di-

agnosis of a student’s problem in order to be able to

provide some personal guidance that will clarify all

questions of the student. Our tool will come to add

and not replace the guidance of a teacher, by offering

a smart educational environment, where the student

can be aware of their strength, weaknesses and the

goal of their level of education (grade). At the same

time, the teacher or tutor can provide specialised as-

sistance with rich content based on a student’s pre-

vious learning path and knowledge gaps. Given that

the tool can classify the exercises and select them in

order to close a student’s knowledge gaps, it should

help to increase the success rates on the ﬁnal exams

of each grade and diploma examinations. The goal

is to develop a recommendation system which, based

on the assessments and the existing rich content, will

gradually follow up the student’s skills until they have

reached the proper level to face difﬁcult exam ques-

tions with conﬁdence instead of fear and embarrass-

ment. The beneﬁts of such a system are of ﬁnancial

and pedagogical nature, such as the standard of qual-

Smart Learning Tool Backend

Learning Path Encodings

Knowledge Graph

Assessments

Figure 6: Smart learning tool backend components.

ity, the re-usability, scalability, convenience and time-

saving.

Our tool could be used as an integrated application

for already existing educational programs. It might

further serve as an indicator in the application pro-

cesses of teachers as well as other job candidates.

In teaching, the tool could potentially be used as a

system for examinations or real-time interaction with

enriched class content, which could provide mixed

media interfaces and allow more interactive learning

based on each student’s needs.

3 POTENTIAL APPLICATIONS

The envisioned technology-enhanced smart learning

environment is scalable and can be used for any ed-

ucational material and domains such as mathematics,

physics, biology and history. Based on our proposed

approach, we could potentially detect the best prac-

tices in learning paths, curricula and exam assess-

ments to instruct a more successful next generation

of students, who are better prepared for exams and

have been individually assisted based on their knowl-

edge gaps. The results of the proposed research might

also encourage the mobility of families and students,

as educators will be able to identify the learning path

of each student. Note that this is in line with poli-

cies towards a modernised and adaptive educational

system, which try to address inequality and unqual-

iﬁed dropouts and a modernised educational policy

that can determine the qualiﬁcations of a successful

education system. Another application might be in

the domain of the screening of applications for en-

rolment at universities or job vacancies. In this case,

the smart learning tool could compare a candidate’s

proﬁle to the criteria of a given position, identify the

candidate’s chances for success and assist them to im-

prove their proﬁle to match the expected level of ex-

pertise.

Furthermore, companies working in the domain of

e-learning solutions might beneﬁt from our research

as they will be able to enhance their products with

a rich model that will help them to share their ma-

terial. Note that the proposed research might also

A Technology-enhanced Smart Learning Environment based on the Combination of Knowledge Graphs and Learning Paths

467

beneﬁt companies who are offering courses to assist

high school education and companies that are par-

tially working on e-learning projects.

Moreover, there might be an economic impact on

companies that offer their employees internal training

and educational growth or learning paths, as they will

be able to provide more personalised trainings and

seminars that will reduce the lost time for overqual-

iﬁed employees, save money for trainings within or-

ganisations, and easily and quickly verify a candi-

date’s knowledge level for a speciﬁc position.

Future educators will be able to provide spe-

cialised individual support to learners, which will in-

crease their productivity and satisfaction. On the

same page, future counsellors will be able to modify

curricula and educational systems in an easy, fast and,

most importantly, accurate way to adapt the needs of

a challenging and constantly improving the academic

and professional environment. Last but not least, stu-

dents will enjoy personalised assistance that identi-

ﬁes their weak points and potential knowledge gaps,

and get to know where they need help in order to

reach their educational goals. Of course, they will

also learn about their strong points, which will further

push them towards pursuing their talents.

4 CONCLUSIONS

We have presented our vision of a smart learning tool

that might support students as well as teachers in their

daily life. The smart learning tool is going to be based

on a knowledge graph which connects the different

knowledge topics and also enables the tracking of

each student’s learning trajectory, making it easier to

identify a student’s knowledge gaps and plan the next

steps for a successful learning trajectory. Further, the

proposed approach helps in detecting good practices

in learning trajectories (learning paths) and method-

ologies, such as which topic should be taught ﬁrst and

based on which methodology. On the practical level,

the smart learning application foresees a large pool

of assessments, which are categorised by the knowl-

edge they require in order to be addressed. This aids

learners to have a more accurate recommendation of

assessments based on their needs and proﬁle. There-

fore, the proposed technology-enhanced smart learn-

ing environment combining knowledge graphs and

learning paths is also of great help for teachers, as it

makes the detection of a student’s knowledge gaps a

more efﬁcient and accurate process and provides them

technology-based assistance for the personalised rec-

ommendation of assessments.

REFERENCES

Cai, J. and Nie, B. (2007). Problem Solving in Chinese

Mathematics Education: Research and Practice. ZDM

Mathematics Education, 39.

Chen, P., Lu, Y., Zheng, V. W., Chen, X., and Li, X. (2018a).

An Automatic Knowledge Graph Construction Sys-

tem for K-12 Education. In Proceedings of L@S 2018,

Annual ACM Conference on Learning at Scale, Lon-

don, UK.

Chen, P., Lu, Y., Zheng, V. W., Chen, X., and Yang, B.

(2018b). KnowEdu: A System to Construct Knowl-

edge Graph for Education. IEEE Access, 6.

Council, N. R. (2000). How People Learn. National

Academies Press.

Doignon, J.-P. and Falmagne, J.-C. (1999). Knowledge

Spaces. Springer Verlag.

Harjula, M. (2008). Mathematics Exercise System with Au-

tomatic Assessment. Master’s thesis, Helsinki Univer-

sity of Technology.

Ilkou, E. (2019). EduKnow: A Framework for Structuring

Educational Material. Master’s thesis, Vrije Univer-

siteit Brussel.

Le, N.-T. and Pinkwart, N. (2014). Towards a Classiﬁcation

for Programming Exercises. In Proceedings of the 2nd

Workshop on AI-supported Education for Computer

Science, Honolulu, USA.

Lecailliez, L., Flanagan, B., and Ogata, H. (2019). Con-

struction of a Knowledge Map-based System for Per-

sonalized Second Language Learning. IPSJ SIG Tech-

nical Report, 2019-CLE-27(3).

Levav-Waynberg, A. and Leikin, R. (2006). Solving Prob-

lems in Different Ways: Teachers’ Knowledge Situ-

ated in Practice. In Proceedings of the International

Group for the Psychology of Mathematics Education,

Prague, Czech Republic.

Linn, R. L., Baker, E. L., and Dunbar, S. B. (1991).

Complex, Performance-based Assessment: Expecta-

tions and Validation Criteria. Educational Researcher,

20(8).

Rizun, M. (2019). Knowledge Graph Application in Educa-

tion: A Literature Review. Acta Universitatis Lodzien-

sis. Folia Oeconomica, 3(342).

Signer, B. and Norrie, M. C. (2007). As We May Link:

A General Metamodel for Hypermedia Systems. In

Proceedings of ER 2007, International Conference on

Conceptual Modeling, Auckland, New Zealand.

South, S. J., Haynie, D. L., and Bose, S. (2007). Student

Mobility and School Dropout. Social Science Re-

search, 36(1).

Stapel, M., Zheng, Z., and Pinkwart, N. (2016). An Ensem-

ble Method to Predict Student Performance in an On-

line Math Learning Environment. International Edu-

cational Data Mining Society.

Zeljko

Sundov and Gregori

c, M. (2014). How to Maximize

and Measure Learning’s Contribution to Business Re-

sults. Journal of Education for Entrepreneurship, 4.

CSEDU 2020 - 12th International Conference on Computer Supported Education

468