AdeLE (Adaptive e-Learning with Eye-Tracking):
Theoretical Background, System Architecture and Application Scenarios
(Extended version of the paper published in the proceedings of the I-KNOW ‘04, Graz, Austria, 2004)
Christian Gütl 1,3 [cguetl@iicm.edu and cguetl@acm.org], Maja Pivec 2,4 [maja.pivec@fh-joannuem.at and
mpivec@izit.si], Christian Trummer 2 [christian.trummer@fh-joannuem.at], Victor Manuel García-Barrios 1,5
[vgarcia@iicm.edu and victor.barrios@campus02.at], Felix Mödritscher 1,5 [fmoedrit@iicm.edu and
felix.moedritscher@campus02.at], Juergen Pripfl2 [juergen.pripfl@fh-joanneum.at], Martin Umgeher2
[martin.umgeher@fh-joanneum.at]

(1) Institute for Information Systems and Computer Media (IICM), Faculty of Computer Science, Graz University of
Technology, Austria [http://www.iicm.edu]
(2) Department of Information Design, University of Applied Sciences (FH JOANNEUM), Graz, Austria
[http://www.fh-joanneum.a]
(3) Infodelio Information Systems [http://www.infodelio.com], Guetl IT R&C [http://www.guetl.com]
and Internet Studio-Isser [http://www.studio-isser.com]
(4) Institute for Information Technology (IZIT), Maribor, Slovenia [http://www.izit.si]
(5) Degree Program in IT and IT-Marketing at Graz University of Applied Sciences CAMPUS 02, Graz, Austria
[http://www.campus02.at]
Abstract
Due to the rapidly growing amount of knowledge, a stronger need emerges for efficient and improved
knowledge acquisition strategies. E-learning can be very helpful for different learning activities in various
learning environments. However, in order to support different teaching and learning paradigms, e-learning
should deal with more than simply reading online lessons. Therefore, content as well as communication and
collaboration have to be supported in a highly personalised manner by e-learning systems. Though, tracking and
grasping the user behaviour in real time remains the most challenging task to retrieve an appropriate and fine-
grained user profile as well as to provide personalised learning content. In this paper we present AdeLE, a
technology-based solution of an enhanced adaptive e-learning framework, which comprises novel solution
approaches for fine-grained user profiles by exploiting real time eye-tracking and content-tracking analysis as
well as a dynamic background library. Based on the global objectives of an enhanced e-learning environment,
the system architecture of AdeLE and the methods used in order to gain fine-grained user information by real
time eye-tracking are addressed. Furthermore, various scenarios in different application domains are illustrated.
Keywords
e-learning, adaptive knowledge transfer, dynamic background library, user profiling, eye tracking, AdeLE
1. Introduction
It is well documented in renowned research publications and studies that the increasing amount of knowledge
and therefore the serious demands on knowledge acquisition for students and employees requires improved and
efficient learning activities. Furthermore, people have to be supported and assisted during their life-long learning
activities. Traditional learning methods and environments do not meet the contemporary needs of our
information society any more. Thus, technology-based solutions have been increasingly established to overcome
these problems. As a result, e-learning can be identified as one of the emerging areas in the last few years, as
shown by means of concrete numbers in the (IDC 2003) study. About 934 Mio US$ were invested in e-learning
worldwide in the year 2003 and the European market is meant to be the best one. Though, a lot of failures and
only a few - in most cases locally restricted - success stories can be identified (see also Baumgartner 2003).
In this paper, we introduce our new solution approach, named AdeLE, which is the abbreviation for
‘Adaptive e-Learning with Eye-Tracking’ and comprises a technology-based solution in the field of adaptive e-
learning exploiting novel methods of resolution for fine-grained user profiling based on real-time eye-tracking
and content tracking information. By applying fine-grained user profiles and real-time behavioural data about
users, a highly personalised information provision and a wide range of new applications are facilitated. Given
that the objectives of technology-based educational environments and their impact on individuals are linked with

complex and context-dependent constraints and conditions, we are conscious that the solution framework
presented in this paper may only be valid for specific sub-fields of e-learning.
In concrete, this paper represents an extended version of ‘AdeLE: A Framework for Adaptive E-Learning
through Eye Tracking’, presented at the IKNOW 04 (see García-Barrios et al. 2004), whereupon the scope of
this version is extended by and focused onto a practical discussion of real-time eye-tracking and a description of
practical application domains. Thus, the reminder of the paper is organised as follows: Section 2. Motivation,
describes key issues and requirements for the proposed system, in Section 3. Related Work, a selection of related
research topics is depicted, Section 4. Real-time Eye-tracking, discusses specific features of eye-tracking
technology and the application of its real-time application for the AdeLE framework, Section 5. Architectural
Overview of the AdeLE Framework, provides a technical insight into the modules of the AdeLE architecture, and
in Section 6. Application Scenarios, examples of application areas for our solution approach are discussed.
2. Motivation
Our previous experiences in the subject of e-learning (e.g. Dietinger 2003; García-Barrios et al. 2002; Pivec
2000) have shown that learners tend to stick to distinct learning methods and teachers favour various teaching
methods. Consequently, e-learning involves more than simply reading online lessons.
Adaptive E-learning
E-learning is a large and complex field of research encompassing a variety of learning and teaching
paradigms, such as constructivistic, serial, symmetric (Jain et al. 2002), cognitive, face-to-face, discovery, and
managed learning (Lennon and Maurer 2003). In addition, various levels of pre-knowledge as well as progress
and difficulties on lessons for individual learners have to be considered. As stated in (Jain et al. 2002), e-
learning, even if standardised, tends to produce asymmetrical learning, as its tools reach out to a dispersed
audience where individuals may arrive at different stages at different times, even if along a common learning
trajectory. Furthermore, the personalisation of instructions and interactivity improves the knowledge acquisition
process. For example, a learner in a classroom setting asks significantly less questions, whereas in an individual
tutoring setting, a learner may ask or is required to answer a lot of questions in any learning session. Thus, the
achievement of individually tutored learners’ performance, as measured by test scores, may significantly exceed
that of classroom colleagues; for details see (ADL 2001) and (Bloom 1984). These are some of the reasons for
considering personalisation and adaptivity to be one of the key issues in modern e-learning environments.
Key Issues and Objectives
Following the observations and findings stated so far, it is worth mentioning that in order to offer the
appropriate learning method and to compile the relevant learning assets and learning sequences for any
individual learner, the gathering and analysis of fine-grained user profiling information is a serious technological
challenge. Furthermore, just providing the technical solution of delivering learning content from static learning
repositories will not meet all requirements for supplying teachers and learners with personalised and timely
information. In particular, the knowledge transfer process defines the key issue within the context of technology-
based instructional environments and it can be interpreted as a holistic phenomenon composed of two related
streams: the teaching process (knowledge generation and delivery as well as assessment of knowledge
acquisition) and the learning process (knowledge acquisition). Both of them have to be supported by a future-
oriented approach.
Thus, the main objectives of an innovative solution for an e-learning system are:
(1) personalised retrieval, management and presentation of relevant and timely information for learning
activities,
(2) support of various learning and teaching paradigms, and
(3) improved knowledge of the users' behaviour in the field of human-computer interaction in general as
well as related to the displayed learning contents in order to gain new insights and input for (1) and
(2) as well.
Considering the above depicted aspects, we believe that a more extensive solution framework is needed,
which allows the binding together of effective modern technologies and solution approaches in order to enhance
the adaptation of knowledge provisioning and to increase the effectiveness of personalisation. This notion builds
the basis of our solution approach, the AdeLE framework. Unlike common methodologies of tracking page
views and mouse clicks, we advocate a combination of fine-grained real time eye-tracking and content-tracking
operations for the user profiling as well as complementing the data stream by interactive system-user dialogs and
online progress testing. Furthermore, in the field of information acquisition for the teaching and learning
activities, we propose besides static learning assets also a highly dynamic, task-specific and personalised
information retrieval component, which we call dynamic background library.
Going the right way?
To conclude this chapter, let us state that we are conscious and aware of the fact that personalisation and
adaptation should not be considered as universal goals or solutions for effective e-learning. E-learning

environments, adaptive systems and personalisation techniques have a lot of proven weaknesses and have been
(often justly) criticised by experts of different fields.
Although we think that it is not necessary to repeat all drawbacks and problems of personalisation
extensively in this paper, consider the following few examples of critical aspects regarding the application of
personalised e-learning. From the knowledge delivery viewpoint, there are some cases where teachers want
students to use exactly the same learning material in order to provide a common experience and thus, to foster
community-based learning. From the viewpoint of knowledge acquisition and the support of different learning
styles as well as considering some findings coming from other research fields, developers of adaptive systems
should (a) exercise care in not allowing an information overload on screen in order to improve the effectiveness
of working memory (for details about cognitive load theory see e.g. (Cooper 1998; Feinberg and Murphy 2000),
(b) enable the scrutability of the user profile in order to give learners the overview and control of adaptational
parameters, and (c) consider privacy and security issues e.g. due to the intrusive character of devices such as an
eye-tracker system (see also Czarkowski and Kay 2003).
3. Related Research Fields
The purpose of this chapter is, on the one hand, to show that the ideas behind personalised e-learning are not as
new as supposed. On the other hand, this fact underscores that the field is broad and has a high potential of
interest and for research. Thus, in accordance to the basic issues behind the idea of the AdeLE solution
framework, a selection of some interesting subjects related to the proposed system is discussed as follows.
Adaptive E-learning, Standardisation and Learning Repositories
The utilisation of adaptive multimedia systems as improved learning environments are well documented in
research work (Brusilovsky 1998a; Brusilovsky 1998b; Hothi et al. 1998; Seeberg 2003), implementation
concepts and systems can be found, for example in (Beaumont et al. 1995; Hockemeyer 1997; Boyle et al. 1998;
ARIADNE 2003; García-Barrios 2001b). Tracking the behaviour of users and analysing their learning progress
are not new research issues, but were demonstrated in classic systems already in the 1950’s. Well known early
systems are CLASS and PLATO (Crowell 1967; Modesitt 1974). In the present, novel user modelling techniques
are important as they allow systems to personalise the human-system interaction processes (Conlan et al. 2002).
Well-defined learner model standards and specifications, like PAPI (Public and Private Information for
Learner - IEEE), IMS LIP (IMS Learner Information Package) or GESTALT (Getting Educational Systems
Talking Across Leading-edge Technologies) already exist (PAPI 2000; IMS LIP 2001; Gestalt 1999).
Furthermore, e-learning capabilities that enable interoperability, accessibility, structuring, packaging and
reusability of Web-based learning content are also finding standardisation through advanced specification
initiatives, such as SCORM (Sharable Content Object Reference Model). Also regarding courseware objects
other well-known standards find broad application, as for example LOM (Learning Object Metadata) in order to
support semantic enrichment, or IMS Simple Sequencing in order to define conditional interrelationships of
different successive nodes along an adaptable learning path (LOM 2002; IMS 2003).
The examples shown in the previous paragraphs point out that adaptivity and personalisation constitute
broad research fields with a relatively long history and a large number of significant results (see also Brusilovsky
1998b; Hothi et al. 1998; Beaumont et al. 1995). However, much work is still to be done towards existing or
emerging issues and challenges, especially more generic solutions are needed.
Further, a lot of work is already done regarding the topic of learning repositories. In the field of static
learning assets there exist various initiatives as listed in (NLII 2004), web information retrieval for assisting
learning activities are discussed for example in (Liaw et al. 2003), and the novel idea of the dynamic background
libraries is well documented in (Dietinger et al. 1999) and (Guetl 2002). Though, to our knowledge no
integrative solution combining static and dynamic learning content in e-learning systems is available so far.
Eye-Tracking Research Field
The field of eye-tracking research is indeed an old but at present a very active discipline. Nowadays, eye-
and gaze-tracking systems are found in several research fields, like scan-path testing on the WWW (Josephson et
al. 2002), research in the area of aviation (Merchant 2001) or of driving environments (Hayhoe et al. 2002),
computational studies about visual cognition (see Rao et al. 1997; Zhai 2003), design implications in web search
tasks (Goldberg et al. 2002), human computer interaction (Ivory et al. 2001), and many others. Quite recently
eye-tracking vendors began to implement real time eye-tracking analysis, but there is still a lack of integration in
profiling systems and exploiting the data flow for personalised content compilation.
To sum it up, let us state that according to our extensive literature survey in the context of adaptive e-
learning, (a) great research work has been done, and (b) several prototype solutions, tools and systems already
exist. However, these are just isolated solutions, which cannot entirely satisfy the teachers’ and learners’ needs in
order to improve a personalised semantic knowledge transfer. Thus, the AdeLE framework, which will be
introduced in following sections, covers new ideas by means of providing a modular and flexible architecture for
adaptive multimedia learning systems in order to enhance the adaptive semantic knowledge transfer process as
well as by combining real time eye-tracking and content-tracking in order to enhance fine-grained user profiling.

4. Real-time Eye-tracking Information
One of the main aims within the scope of the AdeLE project is to observe users’ learning activities in real-time
by monitoring the major possible number of behavioural aspects and personal traits. Learner profile information
of special interest for AdeLE are (a) personal characteristics, such as cognitive or learning styles, (b) momentary
states, like tiredness or mental effort (see García-Barrios et al. 2004), as well as (c) other indicators during the
learning process, such as objects and areas of focus, time spent on objects, frequency of visits, and sequences in
which learning content is consumed (see Preis and Mueller 2003). The advantage of interpreting these data in
real time lies in gaining prompt information about the user’s state. By exploiting eye-tracking data combined
with other user behavioural traits linked with the content provided, a fine-grained learner profile can be tracked
by the system and applied e.g. for personalisation of learning content and navigation. Furthermore, based on this
fine-grained learner information, the target of AdeLE is to gain an insight into the learning strategies that users
apply when using an e-learning platform and to be able to detect patterns indicative of disorientation or other
suboptimal learning strategies. At this point, it should be also mentioned that in order to interpret efficiently
behavioural indicators, it is important to not rely exclusively on eye-tracking data, but to supplement it with
information gained by direct and constant user feedback, i.e. learners must ‘on screen’ not only be informed of
why/how adaptation happens but also get free control of the adaptation procedures.
Eye-Tracking Technology and E-learning
In terms of eye-tracking technology, eye movements, scanning patterns and pupil diameter are indicators of
thought and mental processing involved during visual information extraction (Rayner 1998; Kahneman 1966).
Thus, real-time information of the precise position of gaze and of pupil diameter can be used for supporting and
guiding learners through their learning journey. Very roughly, eye movements can be divided into two
components: fixations, i.e. periods of time with relatively stable eye movements where visual information is
processed, and saccades, which are defined as rapid eye movements that bring a new part of the visual scene into
focus. However, more important indicators can be gained by analysing both components together with other
derived parameters. Finding out reliable parameters is one of the emerging aims of the AdeLE project. To
emphasise the complex situation in this research field consider the following problems and prospects in context.
Gaze duration (i.e. time spent on an object) and fixations are not indicative of attention per se, because one
can also pay attention to objects that do not lie in the centre of the focused region. Nevertheless, by considering
other indicators, such as saccadic velocity, blink velocity and rate as well as eyelid’s degree of openness, a better
and more meaningful approximation can be gained. Saccadic velocity, for example, is said to decrease with
increasing tiredness and to increase with increasing task difficulty (Fritz et al. 1992). Further, blink rate,
decreasing blink velocity and decreasing degree of openness may be indicators for increasing tiredness (Galley
2001). Thus, if tiredness is identified, it should be possible through adaptive e-learning mechanisms to suggest
optimised strategies such as the best time to take a break. At this point it is reasonable to also emphasise that the
user should always retain the final say over whether to accept or reject the system’s suggestions (we call that
controlled adaptation or non-intrusive adaptivity).
AdeLE’s Eye-Tracking System
At the present there basically exist two types of eye-tracking systems on the market: outside-in systems and
inside-out systems. Outside-in systems are characterised by the fact that one or more cameras record the eye of
the participant and trace the gaze in a scene through imaging algorithms. The cameras are positioned in front of
the participant. One of the advantages of these systems is given by the fact that the camera can be integrated into
the monitor, and therefore remains basically invisible (i.e. a relatively non-intrusive monitoring is possible).
Inside-out systems are characterised by a special device that the participant has to wear on the head. The image
of the eye is led into a mini-camera by using mirrors. This mini-camera records the eye and the actual line of
vision is found out through imaging algorithms. More characteristics of both systems along with advantages and
disadvantages related to the requirements of the AdeLE project are outlined in detail in (Pivec et. al 2004). For
further detailed information about general eye-tracker characteristics refer to (Jacob 1995) and (Galley 2001).
For the purposes of the AdeLE project outside-in systems seem to be more suitable, hence they are less
intrusive for the learner and support tracking of the user by regular e-learning lessons. Based on the project
objectives and the requirements for the eye-tracking system, the AdeLE team decided to utilise Tobii 1750.
Figure 1 on the next page shows the utilisation of the eye-tracker system within the AdeLE project, on the left
side a user is being monitored on real-time by the eye-tracker during a learning session and on the right side the
monitored results are examined. The Tobii 1750 is integrated into a 17” TFT monitor and can therefore be used
for many forms of eye-tracking studies with stimuli like Web sites, slide shows, videos and text documents.
Further, the Tobii 1750 does not show any problems with its functional re-acquisition from extreme head-
motions. Another advantage is given by its high tracking quality, i.e. it can be used by young or old people, by
persons with dark or bright eyes, by users with different ethnical-dependent anatomic eye types, by people with
glasses or contact lenses as well as under varying environmental light conditions. More technical advantages of
the system can be summarised as follows: a) high accuracy (0.5 degrees accuracy, bias error), b) compensation
of unparalleled quality of head-motion and drift reduction, and c) binocular tracking with a frequency of 50 Hz.
Further, the system provides a well-designed programming interface with which its automatic functionality can

be configured, enabling no additional manual adjustments of parameters on the device. This interface is utilised
to integrate the eye-tracking system into the AdeLE framework, as described in the next section.



Figure 1: Utilisation of the Tobii 1750 Eye-Tracking system.
5. Architectural Overview of AdeLE Framework
This section gives an overview over the functional and logical architecture of the AdeLE framework, as is
illustrated in Figure 2 below. The main functional components are mainly divided into the following three
logical parts: Core Module, User-centred Modules and Lecturer-centred Modules.
Eye Tracking
Monitor
Content
Tracking
Browser
Interactive
Dialog
Browser
User Information
User Profiling
Database
Collaborative
Filtering
Database
Statistics
Content
Browser
User
Content
Communication
Collaboration
Adaptive Semantic Knowledge Transfer
Content
Browser
Course Ware
Multimedia
Knowledge
Units
Course Topics
Knowledge
Topic Structure
Background
Knowledge
Dynamic
Background
Library
Course Creation and Maintenance
Content
Communication
Collaboration
Course Creator, Lecturer

Figure 2: The architecture of the AdeLE framework.
Core Module
The Adaptive Semantic Knowledge Transfer Module (ASKTM) represents the core module of the
framework. From a global point of view the ASKTM coordinates all the surrounding modules and sends and

requests information to and from them. It compiles pieces of content and meta-information for delivery to the
learners. Separate interfaces are provided for the other two groups of users: course creators (authors) and
lecturers (teachers, trainers or tutors). For media and platform-independence, the information is provided in an
XML schema and can be transformed into various formats. The process of content delivery is depicted in the
upper left and lower left parts of Figure 2.
User-centred Modules
The upper right part of Figure 2 shows the user-centred modules for advanced user profiling. The core
functionality for gaining enhanced and more precise user information is located in the combination of the Eye
Tracking Module (ETM) and the Content Tracking Module (CTM). ETM in combination with CTM provides
real-time fine-grained data regarding the user’s reading and learning behaviour. The ETM also gives the system
hints about concentration, excitement or tiredness of the learner, and consequently, inferring criteria to monitor
efficacy aspects to the knowledge assimilation process.
The entire set of information of user interaction and behaviour is supplied to the User Information Module
(UIM), which is in charge of managing the user modelling and profiling issues. In order to support user-
controlled adaptation, the Interactive Dialog Module (IDM) allows learners to set and change user profile
settings actively. Further, the system also can proactively force user interaction. For example, the latter module
can be used to verify and if necessary adjust any automatically inferred user information. If tiredness is
suspected, the IDM may also be used to suggest a short break or provide a relaxation exercise to the user. We
have chosen the term ‘user’ at the upper-side of the framework, because lecturers may also interact with the
system in order to achieve a direct tutoring intervention in the adaptation processes of the e-learning
environment. The implementation of this feature is intended to allow overriding of automatically generated
system decisions, e.g. to support hybrid learning techniques.
The UIM encompasses three user information databases of different granularity: the User Profiling
Database (UPD), Collaborative Filtering Database (CFD) and Statistics Database (SD). The UPD holds fine-
grained information about a wide range of user interactions (e.g. sequences of scanned and viewed pieces of
information) and more abstracted values of user behaviour types (e.g. level of gained expertise in certain
subtopics). Similar user profiles or user behaviour types are grouped and managed in the CFD (i.e. supporting
stereotyped user-modelling). Through collaborative filtering, the system can proactively suggest particular pieces
of information in proper media by exploiting the collective knowledge of user groups and their behaviour.
Finally, the SD manages abstracted information in a user-independent level. Course creators and administrators
may use valuable information (e.g. identified problematic areas of courseware sections) without violating the
privacy of individual learners. The learning process will be improved, because the system will create or deliver
adapted content by means of tracked statistical data (e.g. by delivering more images/tables for learners that have
problems with large and complicated texts).
Lecturer-centred Modules
The lower-right section of Figure 2 shows the lecturer-centred modules of the AdeLE framework that are
responsible for the course creation process. The Course Creation and Maintenance Module (CCMM) represents
the core module for the entire course management and controls the Courseware Module (CM), the Course
Topics Module (CTM) and the Background Knowledge Module (BKM). Course creators and lecturers can set up
and maintain courses as well as request statistics about their courses. The CM manages pieces of information in
different media types and an extensive set of metadata. This module (CM) can either store pieces of information
locally or just manage metadata and include remotely located sources by caching them.
On the one hand, the CTM manages course content by just defining subsections using meta-descriptors, i.e.
course creators only predefine subtopics and their relations at the time of course generation. During the learning
process, users get dynamically proper and most recent pieces of information out from the pool of the CM. On the
other hand, the CTM permits to manage a course topics structure and a thesaurus for providing automatically
relations between subsections. The BKM dynamically provides additional information within the learning
process and helps course creators to keep pace with most recent information.
Different Viewpoints on the AdeLE Framework
From the point of view of learners, the AdeLE framework provides an adaptive e-learning system with
personalised views of the learning material, i.e. the content is adapted in accordance to pre-knowledge and
learning progress, preferred media types, etc. Furthermore, real time eye-tracking can help to identify areas of
understanding difficulty and enable the provision of selective additional information or explanation. A smart
progress profiling keeps pace with learners’ system interaction and may assist them at further learning sessions.
In addition, learners can get a wide range of dynamic background information.
From the point of view of lecturers, the framework offers a wide range of helpful and smart features for
courseware generation and maintenance. Assuming that courseware modules follow a separation of form and
content, or at least follow a consistent style guide, it allows lecturers to create courseware by simply defining
meta-descriptions of subtopics and their relations. Course authors can also create their own multimedia
knowledge units applicable for course delivery and share them. In the processes of creating new knowledge units
or updating information as well as defining background information for the learners, the dynamic background
library assists the course creators and lecturers. The concept of dynamic background libraries is well documented

in (Guetl 2002) and a prototype implementation has been developed by (García-Barrios 2001b). Statistical
information (e.g. identified courseware pieces with understanding difficulties, subjects of most/weak interest)
supports the maintenance of the courseware.
Based on the architectural overview of the AdeLE framework, a wide variety of applications scenarios are
facilitated. However, in the first stage of the AdeLE project, research work is focused on the utilisation of real
time eye-tracking and content tracking information. In order to get a view on the practical application of the
AdeLE framework in this field of research, some application scenarios are shown in the following section.
6. Application Scenarios
By merging eye-tracking technology with proper content presentation the goal is to identify, evaluate and
develop methods of adaptive instruction for personalised e-learning. Currently, the research efforts of the AdeLE
team concentrate on three application areas, which are discussed in the following paragraphs.
Individual Learning Strategies
The first application scenario deals with the development of methods to extract individual learning strategies
from the learner’s gaze behaviour and adapt against the identified learning style. Comprehensive reviews of
cognitive psychology research indicate that people exhibit significant individual differences in how they learn
(Schmeck 1988; Glaser 1984; Robertson 1985; Honey 1986; Leutner 1998). A simple example being individuals
who have a strong visual memory but weaker verbal processing will find text based material harder to process
than individuals who have stronger verbal skills. In the traditional classroom environment a teacher has the
chance to adapt or explain material to suit individuals’ needs. In e-learning environments where a teacher is
frequently not present, pedagogical material is nowadays more uniformly presented. In this environment
information about the learner’s gaze behaviour would be a great opportunity to optimise material to an
individual’s needs. For example, if somebody prefers text and ignores pictures the amount of pictures presented
could be reduced, and vice versa.
Additional Context Specific Information
The second application scenario is defined by the use of the information gained from the specific content
accessed by the user (specific words, paragraphs, areas of pictures, tables, and the like) in order to provide
additional context specific information. For example, an animated picture could accompany textual information,
whereas the integration of the picture proceeds in relation to the words or paragraphs accessed by the user (see
Figure 3 below and Figure 4 on the next page).

Figure 3: In an e-learning course concerned with Alexander the Great’s Conquest of Persia, a map of
Alexander’s advance in the region is shown. The map content is updated in correspondence to the text
paragraph currently read by the learner. In the example, the second paragraph (“Granikos”) is being read and
the map shows the journey of Alexander from Macedonia to Granikos (green, yellow and red areas indicate
fixations and gaze duration).

Figure 4: The reader has advanced to the fourth and fifth paragraph (“Gaugamela”) and is automatically
shown the corresponding map containing the passage from Issos to Gaugamela.
The AdeLE research team also aims at the possible application of the system in the field of high sensible
knowledge transfer, such as training sessions in nuclear power plants, in the field of aviation techniques or in
military, where it is essential that each section of a content unit has to be read by the learner. By means of the
real-time user behaviour, unseen sections of content units provided to the learner are identified by the system and
again will be supplied to the user at the following content units. Of course, this method can not reflect
information about the knowledge acquisition and about pre-knowledge. However, information about content
sections skipped by the learner are utilised to compile specific assessment tests to check the learner’s knowledge
about these particular concepts.
Appropriate Intervention Strategies
The third interesting application area of the AdeLE research work is based on developing and testing
appropriate intervention strategies when the learner is found to be stuck. The e-learning environment might
intervene when a learner is not focused on a relevant part of the computer screen, or is focused completely
outside the task area for a certain period of time, or the eye gaze is sufficiently quick/jerky for a given period of
time. Just to give one example, in case of knowledge acquisition problems for a particular content section more
detailed content or background information can be provided to the learner.
Other Application Areas
Besides the above-depicted issues, innovative solutions as well as an improved and more profound
understanding are expected in the following areas:
• Improved knowledge of the users' behaviour in the field of human-computer interaction in general as
well as related to the displayed learning contents
• Improved and detailed course progress tracking
• Novel possibilities for identifying the most suitable media and content presentation within knowledge
transfer environments
• Identification of problematic areas in the content flow and/or content structuring
7. Conclusions and Future Work
Evidently, the price of an advanced eye-tracking system plays a decisive role in the application possibilities of
the AdeLE solution approach. Nevertheless, existing systems show that an eye-tracking device can be integrated
into a standard monitor. Due to the continuing trend of rapid technical progress, it may be expected that in the
next few years it would be possible to build a low-cost but high-quality eye-tracking system based on standard
hardware components, which would be suitable for real-time analysis of eye-tracking information as described in
this paper. Some results of the AdeLE project may contribute to find new ways of making advanced adaptive
environments for teaching and learning feasible and affordable for institutions in a relative near future.

Potential target groups that could benefit from the presented ongoing research and proposed innovations
based on eye-tracking supported real-time data capturing and adaptation-based systems are identified as follows:
• Various end-users
Support of 100% knowledge acquisition in application fields such as aviation, traffic, different complex
procedures, risk management, decision support, research on learning, and others.
• E-learning platform and knowledge management platform developers
Inclusion of these innovative approaches and provision of better adaptive/adaptable platforms
• Content publishers
Improvement of content structuring, development of user-centred contents and of contents supporting
various learning stiles.
• Eye-tracking system producers
Development of low cost eye-tracking systems, which are possible to apply in a standard computer
working place.
Acknowledgements
The AdeLE project is partially funded by the Austrian ministries BMVIT and BMBWK, by the FHplus impulse
programme. The support of the following institutions and individuals is gratefully acknowledged: Department of
Information Design, Graz University of Applied Sciences (FH JOANNEUM); Institute for Information Systems
and Computer Media (IICM), Faculty of Computer Science at Graz University of Technology; especially
Hermann Maurer and Karl Stocker. Further information about the AdeLE project can be found at http://adele.fh-
joanneum.at.
References
AdeLE (2005). Adaptive e-Learning with Eye Tracking; project Web site. http://adele.fh-joanneum.at, last visit: 2005-03-01.
ADL (2001). Advanced Distributed Learning: Sharable Content Object Reference Model Version 1.2: The SCORM
Overview, Advanced Distributed Learning, USA, 2001. http://www.adlnet.org/, last visit: 2004-11-20.
ARIADNE (2003). Ariadne Foundation; official website of Ariadne Foundation for the European Knowledge Pool.
http://www.ariadne-eu.org, last visit: 2003-11-28.
Baumgartner, P. (2003). E-Learning an Hochschulen: Didaktik, Modelle, Strategien und Perspektiven; In Proceedings of
OCG e-Future, 2003 (in German).
Beaumont, I., Brusilovsky, P. (1995). Adaptive Educational Hypermedia: From Ideas to Real Systems; in Proceedings of ED-
MEDIA 95, Graz, Austria, 1995, p. 93-98.
Bloom, B. S. (1984). The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One
Tutoring; Educational Researcher, 13 (6), 1984, p. 4-16.
Boyle, C.F., Encarnacion, A.O. (1998). MetaDoc: An Adaptive Hypertext Reading System; Adaptive Hypertext and
Hypermedia; Editors: Brusilovsky, Kobsa, Vassileva; Kluwers Academic Publishers, NL, 1998.
Brusilovsky, P. (1998). Adaptive Educational Systems on the World-Wide-Web: A Review of Available Technologies;
WWW-Based Tutoring Workshop, 4th Conference on Intelligent Tutoring Systems, USA, 1998.
Brusilovsky, P. (1998). Methods and Techniques of Adaptive Hypermedia. Adaptive Hypertext and Hypermedia; Editors:
Brusilovsky, Kobsa, Vassileva; Kluwers Academic Publishers, NL, 1998.
Conlan, O., Dagger, D., Wade, V. (2002). Towards a Standards-based Approach to e-Learning Personalization using
Reusable Learning Objects: E-Learn, 2002, p. 210-217.
Cooper, G. (1998). Research into Cognitive Load Theory and Instructional Design at UNSW; Research report at School of
Education Studies, University of New South Wales, Sydney, Australia, 1998,
http://education.arts.unsw.edu.au/CLT_NET_Aug_97.HTML, last visit: 2005-08-02.
Crowell, F. A., Traegde, S. C. (1967). The role of computers in instructional systems: past and future; Proceedings of the
1967 22nd national conference, Washington, D.C., United States, 1967, p. 417-425.
Czarkowski, M., Kay, J. (2003). Challenges of Scrutable Adaptivity; In Proceedings from the 11th International Conference in
Artificial Intelligence in Education (AIED); Sydney, Australia, 2003.
Dietinger, T. (2003). Aspects of E-Learning Environments: PhD work, Institute for Information Systems and Computer
Media (IICM), Faculty of Computer Science at Graz University of Technology, Austria, 2003.
Dietinger, T., Guetl, C., Knögler, B., Neussl, D., Schmaranz, K. (1999). Dynamic Background Libraries - New
Developments In Distance Education Using HIKS (Hierarchical Interactive Knowledge System), J.UCS 5,1, 1999.
http://www2.iicm.edu/cguetl/papers/dynamicbacklib, last visit: 2004-10-29.

Feinberg, S., Murphy, M. (2000). Applying cognitive load theory to the design of web-based instruction; In Proceedings of
IPCC/SIGDOC '00 (IEEE professional communication society international professional communication conference and
Proceedings of the 18th annual ACM international conference on Computer documentation), IEEE Educational Activities
Department, MA – USA, 2000, p. 353-360.
Fritz, A., Galley, N., Groetzner, Ch. (2002). Zum Zusammenhang von Leistung, Aktivierung und Motivation bei Kindern mit
unterschiedlichen Hirnfunktionsstörungen; Zeitschrift für Neuropsychologie, 1, Heft 1, 1992, 79-92 (in German).
Galley, N. (2001). Physiologische Grundlagen, Meßmethoden und Indikatorfunktion der okulomotorischen Aktivität; In
Frank Rösler (ed.): Enzyklopädie der Psychologie, 4, Grundlagen und Methoden der Psychophysiologie, 2001, 237-315 (in
German).
García-Barrios, V. M.; Gütl, C.; Preis, A.; Andrews, K.; Pivec, M.; Mödritscher, F.; Trummer, C. (2004). AdELE: A
Framework for Adaptive E-Learning through Eye Tracking; in Proceedings of I-KNOW ’04, Graz, Austria, 2004, p. 609-616.
García-Barrios, V. M., Gütl, C., Pivec, M. (2004). Semantic Knowledge Factory: A New Way of Cognition Improvement for
the Knowledge Management Process; Proceedings of SITE2002, Nashville, USA, 2002.
García-Barrios, V. M. (2001). Dynamic Background Library: Learning on demand using Hyperwave eLearning Suite and
xFIND; Project work at Institute for Information Systems and Computer Media (IICM), Graz University of Technology,
Austria, 2001. http://www2.iicm.edu/cguetl/education/projects/vgarcia/Project.pdf, last visit: 2004-11-15.
Gestalt (1999). GESTALT: Getting Educational Systems Talking Across Leading-Edge Technologies; Work Package 3,
D0301 Design and Specification of the RDS (Resource Discovery Service), 1999. http://www.fdgroup.co.uk/gestalt, last
visit: 2004-10-29.
Glaser, R. (1984). Education and Thinking: The Role of Knowledge; American Psychologist 39, p. 93-104, 1984.
Goldberg, J. H., Stimson, M. J., Lewenstein, M., Scott. N., Wichansky, A. M. (2002). Eye Tracking in Web Search Tasks:
Design Implications; Proc. Symposium on ETRA2002, New Orleans, Louisiana, USA, 2002, p. 51-58.
Guetl, C. (2002). Approaches to Modern Knowledge Discovery for the Internet. An Approach to the Information Gathering
and Organizing System xFIND (Extended Framework for INformation Discovery); PhD work at Institute for Information
Processing and Computer supported new Media (IICM), Graz University of Technology, Austria, 2001.
http://www.iicm.edu/thesis/cguetl_diss/diss_cguetl.pdf, last visit: 2003-08-05.
Hockemeyer, C. (1997). RATH, A Relational Adaptive Tutoring Hypertext WWW – Environment; Institute for Psychology
at Karl-Franzens University of Graz, Austria, 1997.
Hayhoe, M. M., Ballard, D. H., Triesch, J., Shinoda, H., Aivar, P., Sullivan, B. (2002). Vision in Natural and Virtual
Environments; Proc. Symposium on ETRA2002, New Orleans, Louisiana, USA (2002), 7-13.
Honey, P. (1986). The Manual of Learning Styles; Maidenhead, Berks, Peter Honey, 1986.
Hothi, J., Hall, W. (1998). An Evaluation of Adapted Hypermedia Techniques Using Static User Modelling; Proceedings of
the 2nd Workshop on Adaptive Hypertext and Hypermedia of the Hypertext'98, Pittsburg, USA, 1998.
IDC (2003). U.S. Corporate and Government eLearning Forecase. 2002-2007; 2003, last visit: 2004-01-29, retrieved from
http://www.idc.com/, last visit: 2004-11-18.
IMS LIP (2001). IMS Learner Information Package (IMS LIP); Version 1.0 IMS Learner Information Package Specification,
IMS Global Learning Consortium Inc., (March 2001). http://www.imsglobal.org/profiles/index.cfm, last visit: 2004-11-05.
IMS (2003). Global Learning Consortium: Overview of Specifications; 2003. http://www.imsglobal.org/overview.cfm, last
visit: 2004-11-04.
Ivory, M. Y., Hearst, M. A. (2001). The State of the Art in Automating Usability Evaluation of User Interfaces; ACM
Computing Surveys (CSUR), ACM Press, 33, 4, 2001, p. 470-516.
Jain, L. C., Howlett, R. J., Ischalkaranje, N. S., Tonfoni, G. (2002). Virtual Environments for Teaching & Learning; World
Scientific Publishing Co. Pte. Ltd.; Jain. L. C. (ed), Series of Innovative Intelligence, Vol. 1, Preface, 2002.
Jacob, R. J. K. (1995). Eye tracking in advanced interface design; In Advanced Interface Design and Virtual Environments,
Oxford University Press, Oxford, 1995, 258-288.
Josephson, S., Holmes, M. E. (2002). Visual Attention to Repeated Internet Images: Testing the Scanpath Theory on the
World Wide Web; Proceedings of the symposium on ETRA2002, New Orleans, Louisiana, 2002, p. 43-49.
Kahneman, D. & Beatty, J. (1966). Pupil diameter and load on memory; Science, 154, p. 1583-1585, 1966.
Lennon, J. ; Maurer, H. (2003). Why it is Difficult to Introduce e-Learning into Schools And Some New Solutions; In Journal
of Universal Computer Science, Vol. 9, No. 10, 2003, p. 1244-1257. www.jucs.org/jucs_9_10/why_it_is_difficult, last visit:
2004-11-10.
Leutner, D.; Plass, J.L. (1998). Measuring Learning Styles with Questionnaires Versus Direct Observation of Preferential
Choice Behaviour in Authentic Learning Situations; The Visualizer/Verbalizer Behaviour Observation Scale (VV-BOS),
Computers in Human Behaviour 14(4), p.p. 543-557, 1998.

Liaw, S.S.; Ting, I.H.; Tsai, Y.C. (2003). Developing a conceptual model for designing a Web assisted information retrieval
system; In Proceedings of the 2003 International Conference on Computer-Assisted Instruction (ICCAI2003).
LOM (2003). Learning Object Metadata; Learning Technology Standards Committee of IEEE, Working Group 12 (WG12),
2002, http://ltsc.ieee.org/wg12, last visit: 2004-11-14.
Merchant, S. (2001). Eye Movement Research in Aviation and Commercially Available Eye Trackers Today; Eye Movement
Summary - Assessing Human Visual Performance, Course at Department of Industrial Engineering, University of Iowa,
USA, 2001. http://arrow.win.ecn.uiowa.edu/56245/FinalEyeTrackingReportAug17.pdf, last visit: 2004-10-25.
Modesitt, K. L. (1974). An excellent mixture for PSI: Computer science, PLATO, knowledge levels; Proceedings of the 1974
annual conference, ACM Press, 1974, p. 89-94.
NLII (2004). Learning Object Repositories; National Learning Infrastructure Initiative, Washington, USA, 2004.
http://elearning.utsa.edu/guides/LO-repositories.htm, last visit 2004-10-25.
PAPI (2000). Public and Private Information; IEEE 2000, Draft Standard for Learning Technology - Public and Private
Information (PAPI) for Learner, IEEE P1484.2/D6, 2000. http://ltsc.ieee.org/, last visit: 2004-11-10.
Pivec, M. (2000). Knowledge Transfer in On-line Learning Environments; PhD work at Graz University of Technology,
Austria, 2000.
Pivec, M., Preis, M. A., García-Barrios, V. M., Gütl, Ch., Müller, H., Trummer, Ch., Mödritscher, F. (2004). Adaptive
Knowledge Transfer in E-Learning Settings on the Basis of Eye Tracking and Dynamic Background Library; EDEN 2004
Annual Conference, Budapest, Hungary, 16-19 June, 2004
Preis, A. M., Mueller, H. (2003). EyeTracking in Usability Research & Consulting: What Do the Eyes Reveal about Websites
& their Users; European Congress of Psychology: Psychology in Dialogue with Related Disciplines, Austria, 2003, p164.
Rao, R. P. N., Zelinsky, G. J., Hayhoe, M. M., Ballard, D. H. (1997). Eye Movements in Visual Congnition: A
Computational Study; Technical Report - National Resource Laboratory for the Study of Brain and Behavior, Department of
Computer Science, University of Rochester, USA, 1997.
Rayner, K. (1998). Eye Movements in Reading and Information Processing; 20 Years of Research. Psychological Bulletin,
124(3): p. 372-422, 1998.
Robertson, I. T. (1985). Human Information-Processing Strategies and Style; Behaviour and Information Technology 4(1), p.
19-29, 1985.
Schmeck, R.R. (1988). Learning Strategies and Learning Styles; New York, Plenum Press, 1988.
Seeberg, C. (2003). Life Long Learning; Modulare Wissensbasen für elektronische Lernumgebungen; Springer, Heidelberg /
New York, 2003.
Zhai, S. (2003). What's in the Eyes for Attentive Input; Communications of the ACM (Association for Computing
Machinery), 46, 3, 2003, p. 34-39. http://doi.acm.org/10.1145/636772.636795, last visit: 2004-10-29.