Considering Human Memory Aspects to

Adapting in Educational Hypermedia?

Peter Ágh and Mária Bieliková

Institute of Informatics and Software Engineering

Faculty of Informatics and Information Technologies
Slovak University of Technology
Ilkovičova 3, 842 16 Bratislava, Slovakia
bielik@fiit.stuba.sk
http://www.fiit.stuba.sk/~bielik

Abstract. In this paper we target the limited capacity of the human

memory while developing adaptive educational hypermedia systems. We
discuss implications of remembering and forgetting for the adaptive hy-
permedia systems development. The forgetting is characterized as a con-
sequence of time passed between two learning events. Knowledge from
psychology is used for stating implications of the human memory prop-
erties for an improvement of the adaptive learning systems. An exper-
imental implementation of the model of remembering and forgetting is
described.

1 Introduction
Current adaptive educational hypermedia (AH) systems recognize several as-
pects of an individual user such as user’s goals/tasks, knowledge, background,
preferences, interests, or user’s individual traits [4]. Important aspect considered
in educational AH systems is undoubtedly a level of the user’s knowledge related
to the learned topic (in the IEEE Personal and Private Information [8] learner
profile denoted as the learning performance). The user model reflects current
state of the user knowledge related to the presented information as it is com-
prehended by the AH system. The user’s characteristics change (evolve) in the
course of learning in accordance with changes of current state of his knowledge
(as evaluated by the AH system).
Most current AH systems assume that the amount of user knowledge only
grows. But increasing knowledge (as a consequence of the remembering) is not
the only process. The user can also lose (e.g., forget) some knowledge. The
remembered knowledge is not stored in the human memory forever but in the
course of time the knowledge can (and some of them will) drop out from the
memory.
?
This work has been supported by the Grant Agency of Slovak Republic grant No.
VG1/ 0162/03 ”Collaborative accessing, analysis and presentation of documents in
Internet environment using modern software tools”.
 Considering mentioned characteristics of the learning process is important
during the learning [10]. We presume that a utilization of the human memory
aspects while developing an educational AH system would also improve the effec-
tiveness of the AH system usage through an improvement of the learning process.
Assume for example the following situation: the adaptive book ”presumes” that
a user possesses adequate knowledge (prerequisites) for understanding a concept
just explained. In spite of truly learned concept some time ago, now – after some
time passed from this learning event – the user forgot some of the previously ac-
quired knowledge (because of long time without any repeating). The knowledge
forgetting causes inconsistencies between the user model as represented in the
AH system (which does not consider the remembering and forgetting in an adap-
tation of the educational material to the individual) and the actual state of the
user’s knowledge. As a result, we will likely observe incorrect recommendation
of the educational AH system.
Described situation occurs due to not considering specific characteristics of
the human memory. In this paper we describe some issues related to the hu-
man memory and implications for adaptive hypermedia. We consider the human
memory as a new aspect of the user’s background modelled in the AH system
user model. We give several suggestions for increasing effectiveness of the AH
system, especially educational AH systems. In the paper we presume some ”min-
imal amount of knowledge” delivered to the user via the AH system because the
effect of the knowledge forgetting process becomes significant with only relatively
large knowledge spaces.
The rest of the paper is organized as follows. In the Section 2, we briefly
present known facts from psychology about the human memory and the pro-
cesses of remembering and forgetting. In the next section, we discuss implica-
tions of the human memory characteristics for adaptive hypermedia and propose
a model, which considers the human memory characteristics. Finally, conclusions
and further directions of our research are stated in the Conclusions.

2 Background of human memory models

The human mind can be viewed as an information processing system. Its archi-
tecture is thought to consist of three basic components: sensory memory, work-
ing memory and long-term memory [2]. These components roughly correspond
to the input (the human mind perceives information from the outside through
the senses), processing (information from the sensory memory is processed in the
working memory) and storage (processed information is stored in the long-term
memory) (see Fig. 1). Naturally, information stored in the long-term memory
can be accessed, or activated to help with the processing in the working memory.
Accessing information is perceived as the remembering that can be viewed as
a usage of the system (to be able to find information later again). This view
provides a useful basis for considering the human memory characteristics during
the learning process [10].
 Fig. 1. Model of human information processing system [2].

Information stored in the working memory can be looked up much faster than
in the long-term memory. The working memory is essential for reading compre-
hension. Frequently, the read sentence is related to the previous sentences, so the
new sentences are considered along previous according their sense. It is believed
that this process is accomplished in the working memory (as a consequence peo-
ple with higher capacity of working memory are able faster understand a text).
One of the most interesting and significant characteristics related to the hu-
man mind is the very small capacity of the working memory known as the magical
number ”seven plus or minus two” [9]. The limited storage capacity of the work-
ing memory is accompanied also by a relatively brief duration (estimates range
from 12 to 30 seconds without a rehearsal), which results in the information loss.
The forgetting is viewed primarily as a consequence of

– fading (trace decay) over time,

– interference (overlaying new information over the old) or
– lack of retrieval cues.

The information loss can be prevented by means of repeating. Here the elabo-
rative rehearsal which in contrast to maintenance rehearsal involves deep seman-
tic processing of a to-be-remembered information item1 is more effective [6]. The
maintenance rehearsal involves only simple rote repetition aiming at lengthening
periods of time the information item is maintained in the working memory. The
elaborative rehearsal can be supported by guidelines.
Accessing an information item can be influenced by several factors. Time of
searching the information item can be cut down with a good guideline. But the
effect of a guideline degrades with the rising number of information items bound
with the guideline. Expectant reason is that the system of guidelines brings a
hierarchical organization of the information items. The benefit is that the search
is performed on the smaller file of information items. However, every new related
information item enlarges the file and aggravates the hierarchy.
1
For example, if an individual is presented with a list of digits for later recall
(4968214), grouping the digits together to form a phone number transforms the
stimuli from a meaningless string of digits to something that has a meaning.
 Mentioned factors imply the forgetting. Function of the volume of remem-
bered information depends on time and has a character of falling an exponential
curve. So called the forgetting curve was first described by Ebbinghaus in 1885
[7]. To test the retention, Ebbinghaus practiced a list of information items until
he was able to repeat the items correctly two times in a row. He then waited
varying lengths of time before testing himself again. The forgetting turned out
to occur most rapidly soon after the end of practice, but the rate of forgetting
slowed as time went on and fewer items could be recalled. It was showed that an
individual lost most of the learned information items in first hours (after 8 hours
was on average remembered less than 40% information items). After this time is
the oblivion less intensive (in average still more than 30% after 24 hours and a
bit lesser than 30% after month). Ebbinghaus also discovered that distributing
learning trials over time is more effective in memorizing than massing practice
into a single session; and he noted that continuing to practice material after the
learning criterion has been reached enhances retention.
For the integrity, let us notice that the information items loss can also have
biological reasons. It is possible that some biological processes necessary for en-
coding, storing or searching are disrupted. For example, in a process of embed-
ding knowledge in the memory some structures of brain including hipocamp and
amygdala are active. Their mutilation has a negative influence on the process of
remembering.

3 Some implications for adaptive hypermedia

While designing an educational adaptive hypermedia system we can take an
advantage of knowing characteristics of both working and long-time memories.
Considering the human memory along other aspects of the student’s background
brings several assets:
– more accurate information about the state of the user’s (student’s) knowl-
edge,
– better aid for remembering the knowledge and automatic repetition of lost
knowledge,
– more effective (adaptive) assistance for students with memory-problems, or
in opposite, assistance for students with more-than-average memory abilities.
First two items are related to the effectiveness of delivery of the educational
material by the AH system in general. The last item enables to accommodate
individual differences in the human memory capacity for personalization.
To achieve mentioned assets several techniques can be used. The most im-
portant are:
– hierarchical organization of the learned material – improves the information
access and enables the effective usage of limited capacity of the working
memory,
– guidelines for effective information searching – helps to overcome an inter-
ference between the information items,
 – forming linkages between the concepts (natural or artificial) – results in a
possibility of elaborative rehearsal and better structuring of the knowledge
space,
– measuring understanding the sense of knowledge (or measure the usage of
the knowledge) – helps in planning of the repeating,
– considering a context of the environment – helps to decrease the amount of
information items considered at the time by giving contextual cues.

Some of the techniques are already used in designing educational adaptive hy-
permedia systems or authoring their content [5].
Following discussion is based on our experiments with the adaptive book
on computer architecture where we applied knowledge on the human memory
characteristics [1]. Our proposals can be easily incorporated to various adaptive
hypermedia educational systems.

3.1 Modeling the process of remembering and forgetting

The simplest method of modeling the remembering and forgetting information

items in the student’s memory is an application of the forgetting curve [7]. Us-
ing only the forgetting curve directly is insufficient because we can only infer
how much per cent from the original grist of the information items has been re-
membered in some point of time. However, we cannot recognize whether specific
information item in a given point of time is remembered or has been lost.
We can say only that the information item (learned at time t and not re-
peated) is remembered with high probability (if according to data about the
user’s memory-losing at time t is remembered more than K% of learned infor-
mation items, e.g. more than 90%) or lost with high probability (likewise).
We have proposed a simple model which reflects the forgetting. It extends
every concept’s traditional performance value from the user model [8] with data
about how much is the knowledge remembered. We call it knowledge activity in
the memory. Every knowledge (represented by a concept in the domain model)
has defined the knowledge activity in the memory represented by a real number.
It’s value must be upon given bound B, otherwise the knowledge represented in
the concept is considered as being lost (from the user’s memory). After a suc-
cessful learning the corresponding concept is set as ”learned” and the knowledge
activity in the memory is set to a value greater than B. Moreover, after every
new user’s session with the AH system, the user model reflects the forgetting
curve by decreasing the knowledge activity for every concept not being used in
the session.
Described approach ensures that the repeated knowledge or knowledge more
used are being lost more slowly. The knowledge-remembering model can be sup-
plemented by including hierarchical binds between the knowledge items in a
domain.
3.2 Remembering and repeating
Knowledge is remembered better if we work more with it. It is not enough if the
information item representing a knowledge only appears many times on a page
presented to the student. A measure of remembering depends on how much is
the knowledge substantial (e.g., whether it is a prerequisite for understanding
another knowledge presented on the page, or whether it is necessary for finding-
out results of exercises) and on the appearance (layout) of the corresponding
information item.
It is advisable to distinguish at least three levels of the rate at which a user
has worked with particular knowledge represented by the information item:
– normal level: the user has worked with the knowledge in such a way that
after the end of the session he has remembered it and can correctly reproduce
it;
– low level: the user has worked with the knowledge less than in the normal
level (e.g., the information item has been mentioned just a few times among
many other information items) and
– high level: the user has worked with the knowledge more than in the normal
level (e.g., the user has intensively and repetitive worked with the informa-
tion item and successfully passed several exercises related to the knowledge).
While in the first case the speed of losing can be computed according to the
standard forgetting curve, in the second case the oblivion is faster and in the
last case slower. Of course, there is no linear relationship: very high measure of
the user’s work with a knowledge does not substantially increase its measure of
the remembering. The measure of remembering of a knowledge item depends on
a ”measure of working” with it. However, the raising is very slow from a certain
level. The reason lays in the memory limits. It is possible to remember more than
at the normal level (e.g., frequently used knowledge, important knowledge) but
not substantially more. On the other hand, if the measure of working has been
low (e.g., the information item has been put down only once) the probability of
remembering the knowledge is very low.
The same is true not only for learned knowledge, but also for the repetition.
After a knowledge has been learned, its activity in the memory in time decreases.
By repeating and using the knowledge, its activity in the memory increases. For
example, if a user studies a page where the knowledge item K is referenced or
repeated, or should be used for understanding other assertions, all these activities
increase the knowledge activity in the memory. Of course, there is also important
measure of the user’s work. For example, if a knowledge was noticed only (in a
text, comment, footnote) or announced, then increasing the knowledge activity
in the memory is futile.
The open issue is the determination of a list of the knowledge items considered
during the inference related to remembering and repeating. It seems that it is
not possible particularize the list automatically. We can count up automatically
the frequency of textual representation of a given knowledge in the given text
but this does not reflect its ”importance”. It may happen that the knowledge
has no textual occurrence in the presented page, however, for understanding the
content of the page the user should work with the knowledge intensively (so it is
repeated many times). On the other hand, another knowledge item may occur
frequently in the text but it is not much important for the comprehension (and
therefore not repeated so much). Information on the mapping of knowledge items
to the content should be provided by he author of the educational material.

3.3 Repetition
Information items the user read on a page are inserted into his working memory.
Because of limited capacity of the working memory the information items are
either moved to the long-term memory or they are lost. To support the process
of moving the information items into the long-term memory (i.e., to enforce the
remembering) it is effective to repeat them.
One possibility is a periodical repetition. After the user has learned given
”amount” of the knowledge, the AH system provides the repetition of the knowl-
edge learned from the previous repetition. The repetition can take several faces.
In our adaptive book it is automatically observed how many new knowledge
items the user has learned. Providing the summation of the occupied items is
greater than the predefined capacity limit the AH system invokes a repetition.
The system generates a page with the resume of learned knowledge (occurrence
of the knowledge items in the information fragments is tagged by the author).
The complexity of the knowledge item is also considered. Described approach
does not give exact results, but it ensures a repetition in time closed to the point
where the user has learned certain amount of the knowledge.
Other techniques of the repetition realized in our adaptive book are:
– repeat at the end of a lesson the knowledge learned in the lesson (final
repetition),
– repeat at the beginning of a new lesson the knowledge learned in the previous
lesson (overall introductory repetition),
– repeat at the beginning of a new lesson the knowledge (assumed) necessary
in this lesson (necessary introductory repetition).
The same can be applied to sessions or various parts of the book content.
Often it is not practical or possible to repeat all of the knowledge items
marked as forgotten. The AH system should select a set of knowledge items for
the repetition. Certain number of the knowledge items is selected and only these
knowledge items are repeated at the beginning of a new lesson. If there is large
number of the lost knowledge items the adaptive book offers a repetition-lesson,
aimed for the repetition only.
Selection can be made on several criterions, for example: random selection,
selection based on time of the acquisition a knowledge (priority is given to the
knowledge acquired longer time ago), selection based on a measure of remem-
bering, i.e. the activity in the memory characteristic is used (priority is given
to the knowledge item with lower activity in the memory), selection based on
prerequisite-dependencies (priority is given to the knowledge item which is sup-
posed to be in the need of the user in the next study time).
3.4 Knowledge space organization

Knowledge space is formed by the concepts (with corresponding information

fragments). The concepts are connected by relations. The currently most used
approaches to structuring the knowledge space are the hierarchical approach and
the network approach [5]. The structure of the hyperspace can aid the repeating
in such a way that the repeating one knowledge item may cause the need of
repeating (in part or in whole) another knowledge item. The same holds for the
forgetting.
For example, if a student is able to compute the volume of a cylinder, he
must be able to compute the square of a number. In opposite, if he has forgotten
how to compute the square of a number, he will not able to compute the volume
of a cylinder. But it is not true that if the student has forgotten to compute
the volume of a cylinder, he also has forgotten how to compute the square of a
number or that if he remembers how to compute the square of a number he also
remembers (and knows) how to compute the volume of a cylinder.
The prerequisite relation is well known relation in adaptive educational hy-
permedia [3]. Considering the human memory characteristics it is useful to dis-
tinguish between domain prerequisites and pedagogical prerequisites. Let A be a
prerequisite of B. If A is a domain prerequisite, the student is constrained in
understanding B with requirement to understand A. If A is a pedagogical pre-
requisite the constraint is weaker and it is possible to comprehend B without
knowing A. As an example, let us present expressions in C programming lan-
guage course. If the adaptive book explains this part using commonalities and
differences between C and Pascal languages, then the knowledge about Pascal
are denoted as pedagogical prerequisites. The student needs a knowledge of Pas-
cal to understand this part of the content. But, when a repetition process is
evoked on the ”expressions in C” knowledge item, the Pascal knowledge item is
not necessary to be repeated.
There can be an objection that the above is not fully true. We may repeat
some topics of Pascal when we repeat C language. For example, some things may
be the same (or similar) and the user may have remembered data like ”in C it is
the same like in Pascal”. The user may also remember the page itself, text or/and
its graphical layout on the page. It is also possible that when he would hear about
some topics of C language, he will bring back some information about Pascal. In
all these cases the user will repeat with some knowledge about C language also
some knowledge about Pascal. This may happen. But after some amount of time
the intensity of repeating related knowledge items will decrease and the user will
repeat only already repeated knowledge and its domain-depended prerequisites.

4 Conclusions

The research discussed in this paper addresses the possibility of improving ef-
fectiveness of learning using adaptive educational hypermedia by considering
the human memory characteristics. Important aspect is limited capacity of the
working memory. We discussed impacts of the human mind nature to the adap-
tive hypermedia systems. Our research is supported by experimental adaptive
web-based book. Known adaptation techniques (annotations of links and con-
ditional inclusion of fragments) are supplemented by an inference based on a
model of the remembering and the forgetting which leads to the repeating. The
base for modelling the remembering is the forgetting curve. The forgetting curve
can be tuned individually for each user which results in more effective repeating
by utilization of individual differences.
We still work on experimental evaluation of issues elaborated in this paper.
Our future work will concentrate on using experiments for proving effectiveness
of the proposed approach. Naturally, we expect that the proposed models should
be tuned for particular usage and differences of the individuals.

References
1. Peter Ágh. Efficient learning using adaptive hypermedia, December 2001. Diploma
Thesis supervised by Mária Bieliková. Slovak University of Technology.
2. Richard Atkinson and Richard M. Shiffrin. Human memory: A proposed system
and its control processes. In K. Spence and J. Spence, editors, The psychology of
learning and motivation. Academic Press, 1968.
3. Paul De Bra, Geert-Jan Houben, and Hongjing Wu. AHAM: A dexter-based ref-
erence model for adaptive hypermedia. In K. Tochtermann, J. Westbomke, U.K.
Wiil, and J. Leggett, editors, Proc. of ACM Conference on Hypertext and Hyper-
media, pages 147–156, Darmstadt, Germany, February 1999.
4. Peter Brusilovsky. Adaptive hypermedia. User Modeling and User-Adapted Inter-
action, 11(1-2):87–100, 2001.
5. Peter Brusilovsky. Developing adaptive educational hypermedia systems: From
design models to authoring tools. In Stephen Blessing Tom Murray and Shaaron
Ainsworth, editors, Authoring Tools for Advanced Technology Learning Environ-
ments. Kluwer, 2003.
6. F.I.M. Craik and R.S. Lockhart. Levels of processing. a framework for memory
research. Journal of Verbal Learning and Verbal Behaviour, (11):671–684, 1972.
7. Hermann Ebbinghaus. Memory. A Contribution to Experimental Psychology.
Reprinted Bristol: Thoemmes Press, 1999, 1885.
8. IEEE. IEEE P1484.2/D7, 2000-11-28. Draft Standard for Learning Tech-
nology. Public and Private Information (PAPI) for Learners. Available at
http://ltsc.ieee.org/, 2000.
9. George A. Miller. The magical number, seven, plus or minus two: sme limits on
our capacity for processing information. Psychological Review, (63):81–97, 1956.
10. Juhani E. Tuovinen. Optimising student cognitive load in computer education. In
Ainslie E. Ellis, editor, Proc. of the Australasian conference on Computing educa-
tion, pages 235–241. ACM Press, 2000.