TOWARDS AUTOMATIC CONSTRUCTION OF ADAPTABLE
COURSEWARE STORYBOARDS
Boyan Bontchev and Dessislava Vassileva
Department of Software Engineering, Sofia University, 5, J. Baurchier blv., Sofia 1164, Bulgaria
Keywords: Automatic Course Construction, Adaptive e-learning, Adaptive Hypermedia Systems, Instructor Tool,
Learning Styles.
Abstract: In last twenty years, researchers have conducted intensive research in the area of principal models, software
architectures and practical system development of adaptive e-learning platforms. Brains are fascinated by
great opportunities for radical improvement of the teaching process by means of applying adaptability at
different levels. There are two general issues of adaptive e-learning – enabling different educational content
delivered to different individuals or groups and, as well, differently formed sequencing and presentation of
that content delivery. This paper presents two approaches for creating and delivering training courses
adaptable to learners with different learning styles. The first one is implemented within a platform for
building edutainment (education plus entertainment) services called ADOPTA (ADaptive technOlogy-
enhanced Platform for eduTAinment). By means of ADOPTA, e-learning courses can be created manually
by an instructor as directed storyboard graphs. Another feasible approach is to generate them automatically
on-the-fly by the adaptive engine. The article discusses advantages and drawbacks of these two approaches
for adaptive e-learning course construction.
1 INTRODUCTION
The hypermedia paradigm is based on usage of
hypertext for organising presentation of structured
content for Internet access. Naturally, such a
paradigm allows introducing models and techniques
for adaptive content delivery. Adaptive Hypermedia
Systems (AHS) make use of them and represent
mainly software applications and platforms for
adaptive e-learning, intelligent tutoring, adaptable
multimedia delivery and adaptive web games. From
the very beginning, AHS try to adapt content in
various ways according the user profile. Bearing that
to the e-learning area, AHS deliver hypertext and
hypermedia content that is consistent with the profile
of individual learner or group of learners (Dagger et
al, 2005). Such a content delivery requires definition
of various pedagogical strategies for a course,
mostly supported by appropriate tools for
instructional design. Each strategy is supposed be
best suited for a particular learner according her/his
learning style, knowledge, preferences and goals
(Bontchev, Vassileva, 2006). Some e-learning
platforms with instructional design tools are as
follows:
 InterBook (Brusilovsky et al., 1996) – it
provides means for the creation and presentation of
adaptive electronic textbooks. Disadvantages of
InterBook are that it does not support advanced
adaptive methods and there is insufficient suitable
interfaces.
 NetCoach (Weber et al., 2001) - knowledge of
each training course is presented as a network of
concepts. However it does not support learning
styles.
 AHA! (De Bra et al., 2006) - learning content is
stored in pages, which are represented as XML files.
The presentation of the content page is determined at
runtime according to predefined conditions. This can
lead to confusion and ambiguity among course
authors.
 ReCourse (Griffiths et al., 2009) – it is not an e-
learning system but rather a tool for creating
learning content in accordance with the IMS
learning design standard. ReCourse provides rich
and user friendly interface, but it supports only IMS
LD.
In this paper, we present an instructor tool, which
covers disadvantages of above examined tools and is
368
Bontchev B. and Vassileva D. (2010).
TOWARDS AUTOMATIC CONSTRUCTION OF ADAPTABLE COURSEWARE STORYBOARDS.
In Proceedings of the 2nd International Conference on Computer Supported Education, pages 368-373
Copyright
c
 SciTePress

integrated within ADOPTA (Bontchev and
Vassileva, 2009) – an ADaptive technOlogy-
enhanced Platform for edutainment, i.e. education
plus entertainment. This instructor tool provides
rich, comfortable and effective interface for creating
courses including various pedagogical strategies.
Moreover our module supports learning styles of all
kinds and is not bound to a specific standard. It is
consistent with our principal adaptability model of
adaptive AHS (Vassileva and Bontchev, 2009) as
described in the next chapter.
Despite the facilities introduced in the instructor
tool, the process of creating adaptive course takes
much times and efforts of an instructor.
Furthermore, not always existing courses can cover
goals of all students. Sometimes a learner with
specific objectives need to pass several courses, part
of the content of which is already known about
her/his. In these cases it is convenient to use
automatic generation of an adaptive course. In this
area there are various successful development such
as PASER (Vrakas et al., 2007), DCG (Vassileva,
1997) and OntAWare (Claus and Holohan, 2009).
All of them are based on domain ontologies and
construct educational content using links between
them and their learning objects. For better
efficiency, very important to them are metadata of
learning objects that give more information when a
particular LO is the most suitable to be used.
The first manual approach for course creating is
implemented as a part of the ADOPTA platform for
building edutainment (education plus entertainment)
services (Vassileva et al., 2009). The second one is
in the process of discussion and planning as future
functionality. ADOPTA is a modular system and
includes: authoring tool for establishing the e-
learning course content, instructor application, and
software engine, which is responsible for adaptable
content delivery to every individual learner.
2 CONCEPTUAL MODEL OF AHS
The ADOPTA platform is based on a newly
proposed hierarchical principle model which tries to
improve the traditional AHAM reference model (De
Bra P. at al, 1999). Table 1 describes the essence of
this model of AHS and provides explanation of its
most important characteristics (Bontchev, Vassileva,
2006).
Table 1: Tabular presentation of the structure of the
conceptual model.
Learner Model - provides description
of the learner character as a triple of
sub-models, namely Goals and
Preferences, Learning Style and
Knowledge and Performance.
Goals and
Preferences
Learning
Style
Knowledge
and
Performance
Domain Model - includes description of
the learning content structure. The
content is granulized in LOs,
interconnected in a ontology of the
knowledge domain. LOs and ontology
are described by metadata (Content
Metadata sub-model) according IEEE
LOM specification and Ontology
Metadata Vocabulary OMV proposal.
Ontology
graph
Learning
objects
Content
Metadata
Adaptation Model - is responsible for
presentation of each course storyboard
as a directed graph (Narrative
Storyboard sub-model), metadata (link
annotations and assessment thresholds)
of each storyboard graph (Narrative
Metadata sub-model) and logic rules for
passing over particular graph
(Storyboard Rules sub-model).
Narrative
Metadata
Narrative
Storyboard
Storyboard
Rules
The model splits the hierarchical structure into two
levels. At first level, the model assures a clear
distinction between Learner, Domain and
Adaptation sub-models. At second level, each one of
these models is divided into three others sub-models.
As shown in table 1, the Learner model describes
profile of each learner such as her/his goals and
preferences, knowledge and performance and
learning styles. For each individual learner
character, the model defines learning style such as
activist, theorist, reflector, or pragmatist or, most
often, as a mix of them. Thus, the learning style can
be polymorphic, as far as the learner usually is not
fixed to a concrete style but rather possesses several
ones, at different level.
The domain model contains structured learning
content. It contains also three sub-models: learning
content as LOs packaged according the SCORM
standard (Díaz, Sicilia, Aedo, 2002), metadata about
LOs and semantic ontologies organizing the content.
The model allows various types of LOs to be used -
narrative content, course tasks, essays, assessment
questions, games, etc. Each one of them could be
associated with one or more narrative content LOs.
The content LOs are created by the author and, next,
they are placed on course pages by the course
instructor.
The adaptation model (AM) takes a central place
in that structure. It contains information about
courses content, semantics of the pedagogical
strategy employed by them and course organization.
Courses are presented by so called narrative
storyboard graphs.
TOWARDS AUTOMATIC CONSTRUCTION OF ADAPTABLE COURSEWARE STORYBOARDS
369

Figure 1: A sample narrative storyboard graph.
Fig. 1 presents a sample for narrative storyboard
course graph. Nodes of a storyboard course graph
are either narrative pages (such as Page 1, Page 2) or
control pages (CP) (such as Control Point 1 and
Control Point 2). Between any two CPs there are so
called work paths (WP) of narrative content pages.
Each one of these content pages is composed of one
or several LOs. For each of these LOs the instructor
can assign a parameter that specifies conditions a
LO to be visible (for example, one such condition
may be test results of a learner in a CP to be over a
certain percentage). Information on these
parameters’ value is used by the adaptation engine in
adaptive content delivery. Moreover the instructor
may define a weight of a WP for each learning style.
Therefore a particular WP may be suitable for one or
several learning styles. The adaptation engine
determines which WP is most appropriate for a
particular learner based on these weight and data
from the Learner model. The control pages are used
for assessment of current knowledge and
performance for a learner, by automatic test
generation. This test is composed of questions
corresponding to the LOs in the pages, which the
learner is visited. The obtained assessment result is
used for update of WP weights.
The conceptual model sketched over proposes
many advantages, especially in assuring strong
independence between learner profile, author
content and pedagogical strategy (Vassileva,
Bontchev, 2009). Moreover, it provides support of
different families of learning styles, content
metadata, and adaptive rule metadata.
3 TRADITIONAL STAGES IN
ADAPTIVE COURSEWARE
DESIGN AND DELIVERY
The traditional workflow of adaptive courseware
design and delivery includes three
main stages as shown in fig. 2:
 Authoring of courseware LOs (usually
organized in domain ontology);
 Instructional design of an adaptive course;
 Adaptive courseware delivery – done in various
ways with different delivery parameters
controlling the adaptation engine.
As far as each of these three phases supplies results
for the next one, it is very important to plan the work
of authors, instructors and supervisors in a coherent
way. In that sense, authors should designed many
domain LOs being of different complexity level and
of various types suitable for any of the learning
styles. Content authors are supposed to do it in order
to provide instructors with e-learning courseware for
constructing various working paths appropriate for
different learner’s characters. As well, instructors
should set appropriate metadata and parameters for
the course pages in order to control courseware
delivery with effective adaptation towards learning
styles and assessment results. As far as this is very
difficult to be obtained in a pure sequential
workflow, transitions from each one of the phases to
another should be allowed.
Figure 2: Stages in traditional adaptive courseware design
and delivery.
3.1 Content Authoring
The authoring process involves content author as a
creator of LOs for a given domain. In many other
approaches, authors of domain LOs have to design a
great number of LOs of various types such as formal
theory, informal LOs, examples, tasks, essay topics,
quests, quizzes, mazes, etc. They have to do that in
order to feed instructors with e-learning courseware
material sufficient for construction of various
working paths for different combinations of learning
styles.
Fig. 3 represents a distribution of LOs types in a
two dimensional space in accordance with their
appropriateness to several learning styles. The plane
is formed by the four learning styles according
Honey and Mumford (Bontchev, Vassileva, 2006).
Within this family of styles, the activist is a
complimentary style to the theorist and, also, the
CSEDU 2010 - 2nd International Conference on Computer Supported Education
370

pragmatist is the opposite style to the reflector. We
have disposed various types of learning objects over
the plane according their suitability for a learner
being dominated by a given learning style or by a
combination of two learning styles (the most easy
case). For sure, given learning character may be
composed by all the four styles – in this case,
various types of LOs may be proposed to the learner
as far as they are suitable for any of the styles.
The distribution shown in fig. 3 is a fruit of our
practical experience and does not pretend to be
punctual or validated according instructional theory.
In other words, we would like just to attract readers’
attention to typification of LOs according their
appropriateness to learning styles. Thus, LOs
produced during the authoring phase are used within
instruction design on different working paths within
the narrative storyboard graph, in order to satisfy
learner expectations.
Figure 3: Distribution of LOs types according their
suitability for learning styles (Bontchev, Vassileva, 2009).
3.2 Course Storyboard Design
The instructor tool is a Flex application for creation
via Internet of courses adaptable to different users
with specific learning styles. The courses are
composed in terms of interconnected pages
represented as nodes of a narrative storyboard. The
narrative storyboard graph is to be processed by the
adaptation engine (AE) in order to choose the best
working path for a particular user. Content pages can
be easily modified by drag and drop of available
learning objects. Fig. 4 shows instructors drag action
from learning objects browser where they are
organized in an ontology graph as defined by the
author. In the course graph, there is one terminal
vertex that represents a control page, i.e. course
exam. A course exam is generated automatically by
choosing some of the questions related to the
learning objects shown on pages of the work path
leading to that CP (as far as questions are designed
by the course author and linked to correspondent LO
within the ontology tree). Thus, the instructor is not
responsible for construction of assessment tests. To
tune the course feedback, he/she can adjust CP
thresholds values, i.e. level of assessment results for
passed exam.
Instructor has also the responsibility to annotate
page links and to set page weight parameters for
each of the learning objects for given page. These
page parameters are used for controlling the adaptive
content selection and, therefore, are very important
for tuning the system. The supervisor of AE may
match parameters value to assessment result and,
thus, he/she is able to control appearance of LOs for
any particular learner. If the parameter of a LO
within the page has high value and the learner has
shown high performance at the last CP, this LO
should be viewed to such a learner. Thus, when
learner asks for the next page, adaptive engine may
hide some objects that are not important for this
user. Links annotation labels can be added also by
instructor to influent user’s decision when a
particular user is choosing among several links. If a
learner abandons the work path determined by AE
(by clicking on a link leading to another page
outside of the path), the AE continues tracking pages
the user has passed through giving the user ability to
return back to the path by adding the link “Return
back to the proposed path” to each of the pages.
The instructor uses a Web based client
application developed in Adobe FLEX 3, as a rich
internet application while the server-side of the
application is developed in Java EE. Instructors may
perform any action concerning creation and update
of narrative storyboard including creating courses,
creating pages, filling pages with learning objects,
interconnecting pages, adjusting learning objects
characteristics, setting link annotations, adjusting
exam thresholds, and checking user feedback.
While editing narrative storyboard, the instructor
has the responsibility to annotate page links and to
set page weight parameters for each of the learning
objects population the page. These page parameters
are used for controlling the adaptive content
selection.
The instructor can parameterize the level of
difficulty of a particular learning object. This
parameter provides information to the adaptive
engine whether or not to show a given learning
TOWARDS AUTOMATIC CONSTRUCTION OF ADAPTABLE COURSEWARE STORYBOARDS
371

object to a particular student with shown knowledge
level.
Thus, given work paths created by the instructor
are appropriate for students with pronounced
learning style. For example students, who can be
determined mainly as theorists, will receive content
materials only for this learning style such as
formalizations, generalizations, etc.
Figure 4: View of the instructor tool.
3.3 Adaptive Content Delivery
Adaptive content delivery is controlled by a software
engine assuring adaptability of courseware content.
Line other approaches (Weber, Hans-Christian,
Weibelzahl, 2001), adaptation takes place mainly on
two levels - adaptive content selection and adaptive
navigation:
 Workflow controlling adaptive content
selection – by means of the administration module, it
is possible to configure start/stop of content
adaptation or of navigation adaptation, how many
questions to generate on a CP, and which LOs to be
visible at a given page for learner with given
assessment results. As well, supervisors can use the
module for monitoring to track the effectiveness of
adaptation.
 Workflow controlling adapting navigation –
first at the beginning of a new WP (here the engine
chooses the path of greatest weight (computed by
the engine itself); next, at the end of the current WP
- involving updates of the weights of the traversed
path and determining whether the student can
continue forward or to return to the start of the path.
4 AUTOMATIC CONSTRUCTION
OF STORYBOARD GRAPHS
The opposite approach of constructing storyboard
graphs by instructors is that one of automatic
sequence construction in a dynamic way. In fact, this
approach excludes the instructional design as an
intermediate stage of adaptable courseware
production (fig. 2).
Figure 5: Selection of LOs within an ontology.
For this goal, the learner is supposed to select within
the ontology all the sub-domains he/she likes to
learn. The next figure presents a part of our ontology
of Java EE LOs, where a learner has selected some
of them (shown in dark ovals) simply by pressing
the mouse. A mouse click over a LO selects it and
all its sub-type LOs below so if the learner would
like to select only part of them he or she has to click
over the rest of LOs. E.g., the LO named “JSPs” is
selected but its sub-type LOs “Scripting” and
“Implicit Objects” are not (fig. 5).
After selection of desired LOs, the automatically
generated storyboard for the particular learner will
include the selected LOs from top to down and from
left to right. In fact, main narrative LOs will be
shown to the learner but LOs of other types will be
present only if they are appropriate for this learner
character. The adaptation engine will track again the
shown LOs in terms to generate the final
assessment; however, learners are free to ask the
engine for intermediate tests at any moment.
CSEDU 2010 - 2nd International Conference on Computer Supported Education
372

5 DISCUSSIONS
The paper (supported by the ADOPTA project
funded by the Bulgarian National Science Fund
under agreement no. D002/155) has presented two,
orthogonally opposite approaches for construction of
courseware content adaptable to learner styles of
individual learners. The fist one requires availability
of course instructor, who uses an instructor tool for
constructing storyboard graph of the course. In
particular, the ADOPTA instructor tool allows
instructors to create within the storyboard different
work paths for different learner’s characters, i.e.
characters pertaining to different learning styles. In
such a way, learners who are predominantly
activists, theorists, pragmatists or reflectors, will
receive partially different courseware content
adapted to their personal learning style. This is
achieved by means of adaptive navigation through
the storyboard graph which is controlled by the
ADOPTA adaptation engine. In the same time, LOs
on the pages shown to different learners may vary
according their complexity and achieved individual
results. Thus, there are two important issues to be
pointed out here:
1. The instructor is responsible for setting the
control points and the WP leading from one CP
to another;
2. The instructor selects LOs allocated on pages of
given WP according their type (suitable for
given learner character) and their complexity.
The second approach of automated generation of
storyboards (i.e., automated sequencing) is very
promising, as far as it is much cheaper and faster.
Moreover, it allows learners to state explicitly their
goals by selecting sub-trees on the ontology with
desired sub-domain LOs. As well, learners are not
supposed to make control assessment tests in
predefined control points – instead, they may ask the
adaptation engine to generate assessment questions
at any page of the sequencing. Thus, the automated
generation of storyboards is more promising in terms
of adaptation flexibility. On other side, storyboards
created by instructors follow a pedagogical strategy
and better balance between LOs types and
complexity which makes them obsolete for many
specific cases.
REFERENCES
Bontchev B., Vassileva D., 2009. Adaptive courseware
design based on learner character, Proc. of Int. Conf.
on Interactive Computer Aided Learning (ICL2009),
23-25 Sept., 2009, Villach, Austria, pp.724-731.
Brusilovsky, P., Schwarz, E., Weber, G., 1996. A Tool for
Developing Adaptive Electronic Textbooks on WWW,
Proc. of WebNet’96 - World Conf. of the Web Society,
October 16-19, 1996, AACE, pp. 64-69.
Claus, P., Holohan, E. 2009. Applications of Semantic
Web Technology to Support Learning Content
Development, Interdisciplinary Journal of E-Learning
and Learning Objects, Volume 5, 2009, pp. 1-25.
Dagger, D., Wade, V. & Conlan, O., 2005. Personalization
for All: Making Adaptive Course Composition Easy,
Special issue of the Educational Technology and
Society Journal, IEEE IFETS.
De Bra, P., Smits, D., Stash, N., 2006. Creating and
Delivering Adaptive Courses with AHA!, Proc. of EC-
TEL, Springer LNCS 4227, pp. 21-33.
De Bra P. at al, 1999. AHAM: A Dexter-based Reference
Model for adaptive Hypermedia, Proc. of the ACM
Conf. on Hypertext and Hypermedia, pp. 147-156.
Díaz, P., Sicilia, M.A. and Aedo, I., 2002. Evaluation of
Hypermedia Educational Systems: Criteria and
Imperfect Measures, Proc. of the Int. Conf. on
Computers in Education, USA, 2002, pp. 621-626.
Griffiths, D., Beauvoir, P., Liber, O., Barrett-Baxendale,
M., 2009. From Reload to ReCourse: learning from
IMS Learning Design implementations, Distance
Education, Vol. 30, Issue 2, pp. 201 - 222.
Vassileva D., Bontchev B., 2006. Self adaptive
hypermedia navigation based on learner model
characters, Proc. of IADAT-e2006, Barcelona, Spain,
2006, pp.46-52.
Vassileva D., Bontchev B., Chavkova, B., Mitev V. 2009.
Software Construction of an Authoring Tool for
Adaptive E-learning Platform, Proc. of BCI’2009,
Thessaloniki, Greece, pp.187-192.
Vassileva D., Bontchev B., 2009. Adaptation engine
construction based on formal rules, Proc. of CSEDU
2009, ISBN 978-989-8111-82-1, Vol.1, Lisbon,
Portugal, March 23-26, 2009, pp.327-332.
Vassileva, D., B. Bontchev, S. Grigorov, 2009. Mastering
Adaptive Hypermedia Courseware, Acta
Electrotechnica et Informatica, Vol. 9 (1), pp. 57–62.
Vrakas, D., Tsoumakas, G., Kokkoras, F., Bassiliades, N.
2007. PASER: a curricula synthesis system based on
automated problem solving, Int. J. of Teaching and
Case Studies, ISSN:1749-9151, Vol. 1, pp. 159-170.
Vassileva, J. 1997. Dynamic Course Generation on the
WWW, Proc. of 8
th
World Conf. on AI in Education
(AI-ED97), Knowledge and Media in Learning
Systems, Kobe, Japan.
Weber, G., Hans-Christian, K., and Weibelzahl, S., 2001,
Developing Adaptive Internet Based Courses with the
Authoring System NetCoach, Hypermedia: Openness,
Structural Awareness, and Adaptivity, LNAI 2266,
ISBN: 978-3-540-43293-7, pp. 226-238.
TOWARDS AUTOMATIC CONSTRUCTION OF ADAPTABLE COURSEWARE STORYBOARDS
373