Procedia Computer Science

Volume 66, 2015, Pages 711–719

YSC 2015. 4th International Young Scientists Conference on

Computational Science

An Approach of Learning Path Sequencing based on

Revised Bloom’s Taxonomy and Domain Ontologies
with the use of Genetic Algorithms
Vadim Shmelev, Maria Karpova, Alexey Dukhanov
Department of High performance Computing
ITMO University
St. Petersburg, Russian Federation

vad1611@yandex.ru, karpovams@niuitmo.ru, dukhanov@niuitmo.ru

Abstract
With the rapid growth of information technology, we can observe an increase amount of e-learning
systems. Thus, students and teachers are exposed to a wide variety of learning resources and learning
objects. They need to choose courses, lessons, and their sequence in order to obtain desired
knowledge. A technique for developing an individualized learning path that meets user requirements is
described in this paper. The proposed approach utilizes ontology and a revised version of Bloom’s
taxonomy for the purpose of creating learning paths and a genetic algorithm for choosing the most
appropriate.

Keywords: Individualized learning path, genetic algorithm, ontology, Bloom’s taxonomy

1 Introduction
Due to the rapid development of information technology in various fields of science, the number of
learning objects (LOs) (Grunwald S. 2007) is growing fast. There is a large amount of LOs stored in
special repositories; for example, (CAREO n.d.) contains over 4200 records of LO metadata.
Additionally, many metadata records are stored in (MERLOT n.d.), (ARIADNE Foundation n.d.), and
other LO repositories. Thus, students and teachers are exposed to a wide variety of learning resources
and learning objects. Also, they are constrained requirements for the courses and background
knowledge of the students. In this regard, individualizing learning materials pose a challenge to
choose more appropriate LOs and create convenient sequence of LO studying.
It is precisely because of the number of LOs that the problem of compiling course materials while
taking learner requirements into account is difficult—heuristic algorithms are commonly employed as

Selection and peer-review under responsibility of the Scientific Programme Committee of YSC 2015 711

c The Authors. Published by Elsevier B.V.
doi:10.1016/j.procs.2015.11.081
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

a solution. (Aldasht 2010) For instance, ant colony optimization was used in (Yao Jung Yang 2009),
and swarm particle optimization was applied in (Sarath Chandar A.P 2010).
In this paper, we propose a LO sequencing technique which takes into account the aforementioned
features of learning competences. The variety of LOs in online repositories is rather expansive, and the
task of finding suboptimal learning sequences constitutes the problem examined in our research.
Therefore, we used the Genetic Algorithm (GA). To determine the proper order of LOs in a sequence,
we use:
x an ontology to find connections between LOs
x the revised version of Bloom’s taxonomy for assessing the “quality” of connections.
The revised version of Bloom’s taxonomy is widely used in the assessment of students’ skills and
knowledge level both before a course and after its completion. The ontology allows us to present
knowledge and list domain areas within the structured matter and to avoid ambiguity in the LO’s use
description.

2 Related works
The problem of course sequencing is under consideration during last fifty years. There are
numerous works devoted to this problem, including those (Sarab Al-Muhaideb 2011), (Brusilovsky
2001), and (Atkinson 1966).
Andrea Stebini and Marco Temperini (Sterbini 2010) describe the LECOMPS framework, which
aims to build personalized courses. The framework implements an iterative program. At its first step,
it selects LOs from a repository for inclusion in a convenient path from the learner’s current
knowledge to the target knowledge.
At (Knolmayer 2003) applied an approach similar to MRP (Materials Requirements Planning).
They utilized Gozinto-type computation to determine the set of objects which may be sequenced to
provide the learner with materials in an accessible way. In constructing the sequence, they represent
the data through a graph-based procedure.
In a paper (Sargsyan 2011), a technique for user-adaptable course planning was described. Course
materials were divided into parts, which were used to construct courses. Each part contains a certain
set of key concepts that a learner should absorb. By using the dependency matrix, they defined
dependencies between parts, and with a genetic algorithm they designed a course (a sequence of
teaching units).
In this paper, we attempt to create a new approach of automatic course creating based on the
abovementioned works with the use of skill levels given in the revised version of Bloom’s taxonomy
and the ontology.
Nowadays, as mentioned, the number of LOs is quickly increasing. For better usability, LOs have
some descriptive standards; among these are standards devised by Dublin Core Metadata, SCORM,
IEEE Learning Object Metadata, IMS, and CanCore. LOs described by some standard are usually
stored in special repositories—examples include: CAREO (CAREO n.d.), MERLOT (MERLOT n.d.),
and ARIADNE (ARIADNE Foundation n.d.).
To define relationships between LOs, an ontology is commonly employed. Within computer and
information sciences, an ontology defines a set of objects and relations between them in order to
model a domain of knowledge. (T. Gruber 2009). Ontologies are filling a more specialized role in
Artificial Intelligence, knowledge engineering (T. Gruber 1993) (Uschold 1996) (Gaines 1997),
knowledge representation (Guarino 1995) (Artale 1996) (Sowa 1998.), and knowledge management
and organization (Poli 1996), among other fields. In this paper, a simplified ontology is used to

712
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

identify connections between prerequisite subjects and the outcome subjects. Bloom’s revised
taxonomy is utilized in estimating the quality of these connections (Bloom 1956). Bloom’s taxonomy
is frequently mentioned and applied within the field of education and has been used by administrators,
curriculum planners, researchers, and other educational agents. Lorin Anderson and David Krathwoh
revisited the cognitive domain (Anderson 2001) and it looks like: create, evaluate, analyze, apply,
understand, and remember.

3 Learning path sequencing

Due to the copious amount of LOs, a special system is needed for course planning that considers
the user’s requirements and level of background knowledge; these determine, respectively, the
course’s outcomes and prerequisites. This paper details the use of genetic algorithm, the
implementation of which is described further.
A course is a sequence of LOs, each featuring a list of prerequisites and desired outcomes.
Prerequisites and outcomes constitute a subject and level pair, as described in Bloom’s Taxonomy.
(For example: “linear equations: apply”.) Through the use of these pairings, we can connect LOs into
sequences we use it to assess the resulting sequences. Way, that we asses sequences we use as
objective function of the genetic algorithm.

3.1 Genetic algorithm

As seen in related works, course sequencing poses challenges, and heuristic algorithms are
commonly used to address them. (Aldasht 2010). We use the genetic algorithm, although other
algorithms may be employed toward similar ends—these include ant colony optimization (Yao Jung
Yang 2009), swarm particle optimization (Sarath Chandar A.P 2010), and others. The genetic
algorithm implemented in this paper is not much changed from our previous realization. (Dukhanov
2015). However, some details merit clarification. A course (sequence of LOs) is presented as an
individual with one chromosome, and each gene is a LO. Each gene has a special mark that is
responsible for the use of the object within a sequence. All chromosomes have equal lengths and
contain all LO offerings for the course compilation in a random order. In order to reduce chromosome
length, it is possible to select some LOs from the initial set of LOs prior to GA application. The initial
population consists of random courses.
Crossover. We have utilized a one-point crossing operator described in (Üçoluk 2002). Each of
our chromosomes have the same length and contain the same LOs. It is senseless to have duplicate
LOs in one chromosome, so we ought to avoid placing duplicates within a sequence. The algorithm
described accomplishes this. After this crossing, we expand our population through the addition of two
offspring.
Mutation. In this procedure, we change the state of the gene using or by exchanging it with
another random gene with a certain probability given as a parameter. We do not touch the elite part of
the population.
Selection. There are various ways to conduct the selection (Goldberg 1991). We implemented
tournament selection with the addition of new individuals before selection, while utilizing the elitism
strategy. The elite portion of the population becomes the next generation without any selection having
occurred within it. The remaining individuals are selected by the tournament method until the new
population size is not equal to the size of the population in its initial state.
The value of the fitness function is determined through use of ontology and Bloom’s taxonomy. A
detailed description of this function follows in the next section.

713
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

3.2 Objective function

After describing the ontology and the revised version of Bloom’s taxonomy, we can determine the
objective function of the genetic algorithm. In order to assess an individual (sequence of LOs,) we
apply the iterative process. At the first iteration we seek out LOs within the sequence that will satisfy
the desired course outcomes and save them in a temporary set. An example of this process is presented
in Figure 1. Each temporary set is denoted by different color. This figure shows CP – course’s
prerequisites, CO – course’s outcomes, and LO1..LO7 – learning objects. Arrows between objects
represent the shortest connections (the ܿ‫ ݓ‬value, see (3)).

CP LO1 LO2 LO3 LO4 LO5 LO6 LO7 LO8 LO7 CO

CP LO1 LO2 LO3 LO4 LO5 LO6 LO7 LO8 LO7 CO

CP LO1 LO2 LO3 LO4 LO5 LO6 LO7 LO8 LO7 CO

Figure 1. Example of iterative process. Temporary sets denoted by different colors.

If some course outcome can be satisfied by the given LOs and course prerequisites simultaneously,
we do not add LOs to the set. If course prerequisites do not satisfy course outcomes but several LOs
do, we choose the LO with the best connection. In the next iterations, we will identify LOs that can
satisfy LO prerequisites in the current temporary set and add them to the new temporary set. It must be
noted that the new LOs must be arranged before the LOs in temporary set. This process continues until
a new temporary set is not empty.
When we add LOs to the temporary set, we add the weight of connection to the fitness function
value. When no one LO or course prerequisite is able to satisfy the current prerequisite, we add
penalty for unsatisfied knowledge to the fitness function value, as in formula (1):
Ͳǡ‹…ƒ•‡ͳǢ
‫ ܨ‬ൌ ‫ ܨ‬൅ ൝‹ሺܿ‫ݓ‬ሻ ǡ‹…ƒ•‡ʹǢ (1)
ܷܲ‫ܨ‬ǡ‹…ƒ•‡͵Ǣ
where case 1 – if course prerequisite satisfied LO’s prerequisite or course’s outcome; case 2 – if some
LO’s outcome satisfied LO’s prerequisite or course’s outcome; case 3 – if LO’s prerequisite or
course’s outcome are not satisfied by any outcome or course’s prerequisite; ܿ‫ – ݓ‬connection weight;
‹ሺܿ‫ݓ‬ሻ – minimal of possible ܿ‫ ;ݓ‬and ܷܲ‫ – ܨ‬fine for unsatisfied LO’s prerequisite or course’s
outcome.

714
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

At the end of iterations we add fines for useless LOs within the sequence to fitness value, as in
formula (2):
‫ ܨ‬ൌ ‫ ܨ‬൅ ܷ‫ ܨܱܮ‬൉ ܷ‫ܱܰܮ‬ (2)
Where  – fine for unused LO;  – number of unused LO; 
To assess connection between outcome and prerequisite we use formula (3).
െ݇
ܿ‫ ݓ‬ൌ  (3)
‫ݐ‬ଵ ൉ ȁ‫ܱܤܦ‬ȁ ൅ ‫ݐ‬ଶ ൉ ȁ‫ܤܤܦ‬ȁ
where ‫( ܱܤܦ‬Distance by Ontology) is the distance that is found, such as number of levels in an
ontology. Two subjects of ontology are connected if one of them is the parent of another. If the
distance between subjects of ontology is further than three levels, we have determined that they are not
connected. For instance, the distance between “Android basics” and “Layouts”, presented in Figure 2,
is equal to two levels. The coefficients ‫ݐ‬ଵ and ‫ݐ‬ଶ depend on the types of LOs that can be theoretical
and practical. For practical LOs, it is the more important distance according to Bloom’s taxonomy,
while for theoretical LOs, the distance by ontology carries more weight. The ݇ coefficient set by
expert.

Figure 2. Example of simplified ontology

‫( ܤܤܦ‬Distance by Bloom) determines distance according to the revised version of Bloom’s

taxonomy. This taxonomy consists of these few levels: create, evaluate, analyze, apply, understand,
and remember. By default, the distance between these levels is equal to one, but can also be set
differently by an expert.
Let us present the example of calculating the weight of connection between “Android studio:
understand” and “Android virtual device: create”. By default, the ݇ is equal to 100, ‫ݐ‬ଵ and ‫ݐ‬ଶ to five.
There ‫ݐ‬ଵ is equal to one in the case that the second object is theoretical.
െ݇ െͳͲͲ (4)
ܿ‫ ݓ‬ൌ  ൌ ൌ  െͶǡͷͶͷ
‫ݐ‬ଵ ൉ ȁ‫ܱܤܦ‬ȁ ൅ ‫ݐ‬ଶ ൉ ȁ‫ܤܤܦ‬ȁ ͳ ൉ ȁʹȁ ൅ ͷ ൉ ȁെͶȁ
The aforementioned algorithm is implemented based on an application that is described in our
previous work (Dukhanov 2015). We improved the objective function of the genetic algorithm and
added the ability to choose the type of LO during its creation.

4 Example of course compiling

To test our approach, we created an experimental application that allows us to create LOs, store
them, and create courses from LOs in manual or automatic (by genetic algorithm) mode. We do not
yet have real, extensive sets of LOs. Existing LO repositories do not consider certain data fields,
unlike the metadata that we use in the algorithm. For these reasons, we created 6 LOs that are
presented in Table 1.

715
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

№ Title Prerequisites Outcomes

1 Introduction to Android basics: understand
android programming

2 Android studio Android studio: apply

installing

3 Android first project Android basics: understand Android project structure:

Android studio: apply understand
Android virtual device: apply
4 Views and their Android basics: apply Views: understand
properties Android studio: apply

5 Layouts Views: understand Layouts: understand

6 Dynamic view Android basics: understand Android basics: Analyze

elements changing Views: apply

Table 1. LO list to test and their prerequisites and outcomes

Ontology of these LOs presented in Figure 2.

Table 1 presents results of few launches of system. System generated courses in automatic regime
adhere to the following parameters of GA:
x Iterations number: 20
x Population size: 1000
x Mutation probability: 1%
x Crossover probability: 60%
x Elite part: 30%
x New random individuals in population at each iteration of GA: 10%
Automatic
Manual course
Id Given prerequisites Given outcomes course creating
creating LO order
LO order
1 – Layouts: remember 1, 2, 4, 5 2, 1, 4, 5
2 – Android virtual device: 1, 2, 3 1, 2, 3
evaluate
3 – Views: apply 1, 2, 3, 4 5, 6, 2, 1, 4, 3
Android virtual device:
evaluate
4 Android basics: Views: apply 2, 3, 4 2, 3
remember Android virtual device:
evaluate
Table 2. Comparing automatic and manual cases.

716
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

CP CO

– Android studio Android Views: apply

installing (№ 2) studio: apply

Android basics: 2
remember
Android studio: Android project
apply structure:
Android first understand Android
Android basics: project (№ 3) virtual device:
Android virtual 3
understand evaluate
4 device: apply

Figure 3. Example flow of LO in course sequenced automatically for experiment №4.

As shown in Table 2, the system offers proper (specific) LOs and their order. In the first
experiment, the system offered the course similar to created by humans. The order of first and second
LOs is insignificant because there is no prerequisite LO. The At third experiment offered a somewhat
longer course than what was created manually. There, LOs with numbers 2, 1, 4, and 3 are not
dependent upon each other and can be exchanged. Thus, the proposed sequence contains a manually
created sequence with few redundant LOs. The 4th experiment does not offer LOs for learning
“views”, as this knowledge is related to course prerequisites. Its flow and dependencies are presented
in Figure 3, where CP – course’s prerequisites, and CO – course’s outcomes. The line with number 2
connects the course’s prerequisite and the course’s outcome directly, because these subjects are
adjacent within the ontology. Manually created course have one additional LO, as human agents
deemed it necessary to learn it as well.

5 Conclusion and future works

In this paper, we outlined a technique for course sequencing of LOs. The described technique is
based upon a genetic algorithm. The objective function uses the revised version of Bloom’s taxonomy
and domain ontology. The ontology allows us to determine the connection between LOs, while
Bloom’s taxonomy assists in assessing the quality of this connection. The proposed approach is
implemented based on the application software that is described in our previous work (Dukhanov
2015). The application is capable of sequencing course according to some requirements and
background knowledge. This system assists in courses sequencing from different LOs, for example in
interdisciplinary directions which are typical of courses in master's program in computational science
(Krzhizhanovskaya 2015). In the future, we intend to further tune the algorithm in order to find a more
accurate solution, improve the ontology that is used by the system, and implement the described
system as a public service available online. For online implementation we would like to expand our
repository by adding LOs from other existing repositories. For this purpose, we need to develop a
parser to complement the metadata with data necessary to our algorithms.

717
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

6 Acknowledgment
This work was financially supported by the Russian Science Foundation, Project #14613 "Big data
management for computationally intensive applications"

References
Aldasht, M. M., Saheb, M. H., Najjar, I., Tamimi, M. H., Takruri, T. O. "University course
scheduling using parallel multi-objective evolutionary algorithms." (Journal of Theoretical and
Applied Information Technology) 2010.
Anderson, L.W., Krathwohl, D.R., Airasian, P.W., Cruikshank, K.A., Mayer, R.E., Pintrich, P.R.,
Raths, J., Wittrock, M.C. "A Taxonomy for Learning, Teaching, and Assessing: A revision of Bloom's
Taxonomy of Educational Objectives." 2001.
ARIADNE Foundation. ARIADNE. http://www.ariadne-eu.org/ (accessed 07 27, 2015).
Artale, A., Franconi, E., Guarino, N., and Pazzi, L. "Part-whole relations in object-centered
systems: an overview." Data and Knowledge Engineering 20 (1996).
Atkinson, R.C., Groen, G.J. "Models for optimizing the learning process." (Psychological Bulletin)
66, no. 4 (1966).
Bloom, B.S., Engelhart, M.D., Furst, E.J., Hill, W.H., & Krathwohl, D.R. " Taxonomy of
Educational Objectives – The Classification of Educational Goals ." 1956.
Brusilovsky, P. "Adaptive Hypermedia." (Kluwer Academic Publishers), no. 11 (2001).
CAREO. Campus Alberta Repository of Educational Objects. http://www.careo.org/ (accessed 07
28, 2015).
Dukhanov, A.V., Karpova, M.S., Shmelev, V.A. "Proceedings - Frontiers in Education
Conference, FIE - 2015." An Automation of the Course Design Based on Mathematical Modeling and
Genetic Algorithms. 2015.
Gaines, B. "Editorial: Using Explicit Ontologies in Knowledge-based System Development."
International Journal of Human-Computer Studies, 1997.
Goldberg, D. E., Kalyanmoy, D. "A comparative analysis of selection schemes used in genetic
algorithms." (Foundations of Genetic Algorithms) 1991.
Gruber, T. Ontology. Tom Gruber. 2009. http://tomgruber.org/writing/ontology-definition-
2007.htm (accessed 07 28, 2015).
Gruber, T.R. "A Translation Approach to Portable Ontology Specifications." Knowledge
Acquisition (Knowledge Acquisition) 5, no. 2 (1993).
Grunwald S., Reddy R. "Concept Guide on Reusable Learning Objects with Application to Soil,
Water and Environmental Sciences." ecolearnit.ifas.ufl.edu. 2007.
ecolearnit.ifas.ufl.edu/documentation/concept_guide.pdf (accessed 07 27, 2015).
Guarino, N. "Formal Ontology, Conceptual Analysis and Knowledge Representation."
International Journal of Human-Computer Studies - Special issue: the role of formal ontology in the
information technology 43 (1995).
Knolmayer, G. F. "Proceedings of the 36th Hawaii International Conference on System Sciences."
Decision support models for composing and navigating through e-learning objects. 2003.
Krzhizhanovskaya, V.V., Dukhanov, A.V., Bilyatdinova, A.Z., Boukhanovsky A.V., Sloot, M.A.
"Russian-Dutch double-degree Master’s programme in computational science in the age of global
education." Journal of Computational Science (elsevier) 10 (2015): 288–298.
MERLOT. MERLOT II. http://www.merlot.org/merlot/index.htm (accessed 07 28, 2015).

718
An Approach of Learning Path Sequencing Shmelev, Karpova and Dukhanov

Poli, R. "4th Conference of the International Society of Knowledge Organization (ISKO 96)."
Ontology and Knowledge Organization. Washington, 1996.
Sarab Al-Muhaideb, Mohamed El Bachir Menai. "Evolutionary computation approaches to the
Curriculum Sequencing problem." NATURAL COMPUTING, 2011.
Sarath Chandar A.P, Dheeban S.G, Deepak V., Elias, S. "Sixth International Conference on
Natural Computation (ICNC 2010)." Personalized e-course composition approach using Digital
Pheromones in improved particle swarm optimization. Yantai, Shandong: IEEE, 2010. 2677 - 2681.
Sargsyan, S. G., Hovakimyan, A. S., Barkhudaryan, S. V. "Genetic algorithms in e-learning
systems." (Artificial Intelligence Applications to Business and Engineering Domains) 2011.
Sowa, John E. "Knowledge Representation: Logical, Philosophical, and Computational
Foundations." (PWS Publishing Co) 1998.
Sterbini, A., Temperini, M. "Frontiers in Education Conference (FIE), 2010 IEEE." Selection and
sequencing constraints for personalized courses. Washington, DC: IEEE, 2010.
Üçoluk, Göktürk. "Genetic Algorithm Solution of the TSP Avoiding Special Crossover and
Mutation." (INTELLIGENT AUTOMATION AND SOFT COMPUTING) 3, no. 8 (2002).
Uschold, M., Gruninger, M. "Ontologies: Principles, Methods and Applications." The Knowledge
Engineering Review 11, no. 2 (1996).
Yao Jung Yang, Chuni Wu. "An attribute-based ant colony system for adaptive learning object
recommendation." Expert Systems with Applications (Expert Systems with Applications) 36 (2009):
3034–3047.

719