Volume: 6 Issue: 2 Year: 2009
Student modeling: Recognizing the individual needs of
users in e-learning environments
Sibel Somyürek*
Abstract
Along with numerous universities and large trading companies heavily relying on e-learning
environments to train their students and employees, the design and development of adaptive
educational hypermedia that customize the content and navigation for each student has gained
importance and priority all around the world. This study aims to describe the concept of
student modeling, heart of the adaptive learning systems, and analyze the information
collection, construction and updating phases of a student modeling process. In the study, the
classification of student models in numerous ways is explained, and the different methods
employed in the representation of information in the student model are addressed. Moreover,
the problem of uncertainty, which is one of the most important challenges in the student
modeling process, is mentioned, and the trends in student modeling are discussed.
Keywords: Student modeling, inference, modeling techniques, adaptive educational
hypermedia systems
*
Ins.Dr. Gazi University, Computer Education and Instructional Technologies, ssomyurek@gazi.edu.tr
430
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
modelleme: E-
Sibel Somyürek*
Özet
t
bilgi göst
belirsizlik probleminden
Anahtar Kelimeler:
*
Ögr.Gör.Dr., Gazi Üniversitesi, Bilgi
, ssomyurek@gazi.edu.tr
431
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Introduction
Today, along with the huge rise in the amount of information required to be learned
and the developments in digital technology, numerous universities and large trading
companies all around the world began offer internet-based education to train their students
and employees (Ebner, et al., 1999; De Bra, 1996). Due to the accessibility of internet-based
education environments in terms of number and diversity of education programs, students
who take classes over the Internet tend to be more heterogenously distributed than the
students in the traditional class setting. Therefore, the need to personalize the learning
material for each student becomes of more importance in internet-based education
environment. This problem is attempted to be eliminated by developing adaptive educational
hypermedia (Surjono and Maltby, 2003).
Adaptive educational hypermedia are advanced hypermedia systems that structure the
learning environment by building a model of the student’s goals, interests, and preferences
and customize the education for each student (Brusilovsky, 1998). In educational
hypermedia, according to the prior knowledge of students or the relations between topics, it is
possible make adaptations such as suggesting relevant topics for studying, offering extra
information for those who would like further information, listing down the topics preferred
the most by students, presenting of other subjects about which different students interested in
while studying the current subject, presenting the links in a different order for each students,
etc. In order to provide different services to different users, an adaptive system needs to
define and distinguish a user or user groups (Zhang and Ghorbani, 2007). Without knowing
anything about the user, an educational hypermedia application will offer the same learning
content and pedagogic methods to each student (Fröschl, 2005, p:10). Therefore, a user
model that recognizes the individual learning needs and preferences is the key element for
personalization in adaptive systems.
User Model
A model is an abstract presentation of something that exists in the real world (Koch,
2000, p:35). A user model, on the other hand, is the presentation of a mental status (such as
knowledge, preference, background, and experience) related to a context in the real world
(De Bra, 1999). Stated differently, the user model mean the systems set of beliefs about the
user’s (Kay, 2000).
432
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Figure 1. User Model (Kay, 2000)
In adaptive systems, the user model stores the specific information on each individual,
and in this way enables the system to identify different users (Butz, Hua, and Maquire, 2006).
The entire user model building/updating process is called user modeling. The fundamental
goals of user modeling may be listed as follows (Jameson et al., 1997):
•
To assist the user in locating information;
•
To customize the information presented to the user;
•
To modify an interface according to the user;
•
To select appropriate instructional exercises or interventions;
•
To provide feedback to the user on the level of their knowledge;
•
To reinforce collaboration,
•
To predict future behavior of users.
Since the core task for any user modeling system is predicting future behavior of
users, there exist numerous user models which automatically identify user patterns and give
433
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
suggestions to users on what to do (White, Bailey and Chen, 2009; Bohnert, 2008; Sas et al.,
2003; Barra, Malandrino, Scarano, 2003; Trousse, 2000).
The user in adaptive learning systems is the student. Therefore, in adaptive learning
systems, the concept of “student model” is used instead of user model.
Student Model
One of the most important features that differentiate intelligent learning systems
from traditional learning systems is the ability to interpret student behavior that make
modeling of learning and thinking processes of student possible (Shute and Psotka, 1996).
The student model is one of the four dimensions that intelligent learning systems must
include and student modeling constitutes the essence of researches on intelligent learning
systems such as adaptive systems (Holt et al., 1994).
Classification of Student Models
Student models vary widely and may be classified in different ways depending on
their different features. For instance, student models can be divided in two groups according
to the structure of the information as “field dependent” and “field independent” on the
subject. Information that is field dependent indicate the knowledge and capability level of the
student related to the course content offered within the adaptive learning system, while field
independent information contain information independent of content such as skills,
motivational status, preferences, and learning styles (Brusilovsky, 1994; Shareef and
Kinshuk, 2003).
Another classification is made based on the the information gathering source (single
user – group of users) in the inference mechanism of the student models as “content based”
and “collaborative”. Content-based student models are used in situations where past behavior
of the user may give an idea on his future behavior. For instance, if there is previous
information indicating that a user likes movies such as “Star Wars”, “Raiders of the Lost
Ark”, “Air Force One”, a similar movie such as “The Eyewitness” may be suggested to that
user. On the other hand, collaborative models are used when a user’s behavior is similar to
that of other users. In this approach, the student model is constructed in the light of
information collected from a group of students, and this model is used to make estimations
about one user. For instance, where the favorite movies of a user are similar to those of a
434
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
group of users, other favorite movies of the users in this group may be suggested to that user
(Zukerman and Albrecht, 2001).
Student models may be addressed in two different groups as “visible and opak”. If the
student can see and/or change the student model, the student model is called “visible”;
otherwise, it is called “opaque”. Visibility requires the user interface to ask questions and
gather information about certain features from the student and display a part of or the whole
of the model to the student (Höök, 1998; Kay, 1995). In visible student models, a more
correct model can be constructed with the partnership of the user and the system. However,
the users may want to define themselves as an expert to the system or a playful or curious
user may want to see how system responses would change when she acts differently. So, the
users being able to view the model and make changes may lead to gathering incorrect
information from time to time (Kay, 2000).
Student models may be addressed in two different groups as “dynamic and static”.
Dynamic student models entail obtaining dynamic information about the user based on the
user’s interaction with the system. The student model is constantly updated with the
information obtained in this way. In static student models, the information is obtained from
queries or observations. Information obtained in this way is gathered either the first time the
user uses the system (in the initial phase) or periodically (Koch, 2000, p:38).
Student models may be classified in four groups as “prediction, recommendation,
classification and filtering” based on the type of task for which the model is going to be used.
Prediction is the capability of anticipating future needs of student using past student behavior.
Recommendation is the capability of suggesting interesting elements to a student according to
some extra information not based on the past behavior of the user. Classification builds a
model that classifies items into one of several predefined classes. Filtering is the selection of
subset items from the original set of items to provide each student with information
corresponding to her/his preferences (Frias-Martinez et al., 2005).
Student Modeling Process
The first decision to be made when building a student model is what aspects of the
student should model (Zhou and Evens, 1999). The student model should be constructed with
students’ goals, knowledge, capabilities, preferences, multimedia experience, interests,
435
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
personality in mind, which are effective in their learning (Kobsa, 2001). The student
modeling process can be seen in figure 2.
Figure 2. Student Modeling Process
The student modeling process has three basic phases:
•
Gathering data related to the student characteristics
•
Constructing the student model, and
•
Updating the student model.
Gathering data related to the student characteristics
In order to construct the appropriate adaptations on the adaptive systems, the system
should obtain some information on the user and the user context (Herder, 2006, p:28). To
begin constructing the student model, information about the student (goals, plans, attitudes,
capabilities, knowledge, beliefs of the learner, etc.) should be gathered up and transferred to
this model (Fröschl, 2005, p:36). In the student information gathering process, both static and
dynamic information is collected. Static information, including personality, cognitive style,
etc., is obtained from student records and is quite stable. However, dynamic information that
436
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
shows student performance and improvement throughout the student’s interaction with the
system is dependent on the field and changes in the learning process. Dynamic information is
obtained from the results of the reaction and behavior of the student while using the system
(Alotaiby, 2005).
The methods of gathering data related to the student characteristics may be listed as
three main topics:
a. Direct questions: The initial information needed to construct the student modeling
may be obtained via direct questions to the user. This method is an effective way to
gather general information on the user. It may be used to determine the demographic
data, interests, preferences, etc. of students. Questionnaires, forms, pre-tests,
psychological tests may use in gathering data. However, it is difficult to determine the
right number of questions, and too many questions may disturb the user (Beaumont,
1994; Tsiriga and Virvou, 2003; Nielsen, 1998). In the initialisation of the student
model of existing adaptive systems, self-report measures through questionnaires or
tests are commonly used (Limongelli et al, 2009; Somyürek, 2008; Surjono, 2007;
Triantafyllou, Demetriadis and Pomportsis, 2003).
b. Assumptions: In situations where more information on the student is needed but it
cannot be obtained in any other way, one may make assumptions. For instance, if it is
unknown that the student has background information on the subject, it can be
hypothesized that he has no knowledge on the subject in the beginning. Furthermore,
in this method, similar students may be stereotyped and therefore group-specific
options may be employed. For instance, the system may offer the help option to
inexperienced users visible when they first log in to the system (Fröschl, 2005, p:37;
Koch, 2000, p:47).
c. User-system interaction: The usage information obtained throughout the interaction
of the user with the system is the most important information obtained about the user.
This way, the pages that the user visits, access length and frequency, system log out
time, answers given to the questions in the system, searches conducted, etc. may be
determined. However, information gathered in this way may not be completely
reliable. For instance, the fact that a user browses a page does not mean that she paid
substantial attention to that page (Zhang and Ghorbani, 2007). A typical user exhibits
437
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
patterns when accessing a hypermedia system and the set of interactions of the user
with the system containing those patterns can be stored in a log database in the server.
In this context, data mining and other machine learning techniques make it possible to
to recognize regularities in user trails and to integrate them as part of the user model
(Frias-Martinez, Magoulas, Chen, Macredie, 2005).
The information obtained using the methods above should be transferred to the
variables properly that constitute a student model. Information may exist in three ways in the
variables that make up the student model: boolean, discrete, and continuous. Variables stored
in boolean form may take two values for each situation, right and wrong. For instance, the
situation of a student knowing a concept may be stored as “knows” or “doesn’t know”.
Discretely stored variables may take limited number of values for each situation. For
instance, the situation of a student knowing a concept may be stored and classified as “knows
well”, “knows”, or “doesn’t know”. Variables stored in continuum may take values between
0 and 1 for each situation. For instance, the situation of a student knowing a concept may be
composed of values that vary between 0 and 1. In this case, 0 indicates completely unknown,
1 indicates completely known, and the values between 0 and 1 (0.2, 0.35, 0.82, etc.) indicate
the status of knowing. In an adaptive learning environment, the possibilities for adaptation of
the system to each student are greater in the continuous storage of variables as opposed to
boolean and discrete storage. However, working with continuous variables requires system
developers to deal with more complicated data (De Bra, 1998).
Constructing the student model
When constructing a student model, it is not sufficient to collect information merely
on the student. The collected information should be used to represent the learning process of
a student and to make decisions like determining the path which a student will follow during
his/her learning process or choosing the most adequate pedagogical strategy for the next
subject which will be presented (Wenger, 1987).
This topic focuses on uncertainty problem and student modeling techniques.
Uncertainty problem
One of the most important problems encountered when constructing a student model
is uncertainty (Butz, Hua and Maguire, 2006). In order to construct the student model, it is
438
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
required to obtain some information such as the student’s behaviors’ around the system,
interests, personal characteristics, and in the light of this information, some facts such as the
student’s learning speed and learning preferences should be divulged. However, it is
generally not possible to figure out exact information by using the data to interpret on how
much the student knows on a certain subject, how the student’s learning preferences will
change in the study process, etc., as individual behaviors are complicated, and may change
instantly with situation, environment and time. For instance, a student may prefer to study a
content using captured videos while he may prefer to study a similar content using interactive
examples. Furthermore, the same student may desire to study the same subject in different
ways at different times. Therefore, in the student modeling process, coming up with correct
estimations on the user based on uncertain information is an important problem that should be
addressed (Butz, Hua and Maguire, 2006). The first student modeling systems attempted to
make decisions using manually constructed rules reached by analyzing various situations
about the problem, in order to make deductions from observations on the user. These systems
have not been able to attain the desired performance due to heavy dependence on the source
and not being able to generate solutions under uncertain conditions (Zukerman and Albrecht,
2001). Furthermore, serious problems occurred in the construction and updating of these
systems. First modeling approaches worked with a system based on accepting an absolute
right or wrong, also known as Boolean logic. In Boolean logic, any situation has to be
assigned one of the two values such as "yes" or "no", "1" or "0", or "knows" or "doesn’t
know”. Therefore, no matter how detailed the models constructed using this logic, they do
not completely reflect the truth. The reason is that the real world is complicated, generally
due to uncertainty and the lack of certain judgements (Baykal and Beyan, 2004).
As traditional information presentation methods proved insufficient in modeling
human behavior, the question of how to deal with uncertainty has been a rapidly spreading
and increasing research topic from early 1990’s on (Jameson, 1996, Butz, Hua and Maguire,
2006). In uncertain situations, the systematic pre-planned numeric forecasts can only be made
after certain acceptance and assumptions (Baykal and Beyan, 2004). Due the possibility of
accessing large quantities of electronically available data and by advances in machine
learning of artificial intelligence, the predictive statistical models constructed on these
acceptance and assumptions gained momentum and began to be used in student modeling
(Zukerman and Albrecht, 2001). Artificial neural networks, Demster Shafer approach, and
439
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Bayesian networks may be given as primary examples to these models (Jameson, 1996;
Zukerman and Albrecht, 2001). These models make it possible to construct and update
student models composed of complicated and extensive data, and to make numeric forecasts
on students’ behaviors.
Student modeling techniques
In constructing student models, stereotype model, overlay model, perturbation model,
fuzzy logic and machine learning techniques such as Bayesian network are predominantly
used.
Stereotype model
Stereotype model is a model predominantly used in student modeling due to its ease
of construction. Forming groups based on common characteristics of individuals
(beginner/expert, field dependent/field independent, etc.) constitutes the core of this model
(Figure 3). In the light of initial information gathered, students are assigned to groups
(stereotypes) where they have similar characteristics.
Figure 3. Stereotype Model
People frequently make assumptions based on simple observations (Rich, 1979). For
instance, faced with the knowledge that a person is a judge, it may be deducted that this
person is over the age of forty, has a good education and is well respected. Though not all of
these assumptions are true, unless day-to-day experience proves to the contrary, they are
made. In constructing a stereotype model, similar to daily life assumptions, groups are
formed based on certain assumptions. Stereotype model is convenient in situations where it is
possible to obtain sufficient information on the users. For instance, this model may be used
440
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
when constructing packaged content about university students. Suppose we have evidence
that Bob is a first-year computer science student who uses the piano hall of the university. It
would be appropriate if a page with Computer Science and music events on campus content
appears when he opens up the university web page. In addition to this, it may be expected
that Bob is interested in general culture topics. Also, the information that Bob is not
interested in theatre may be obtained with a questionnaire and added to Bob’s stereotype
model, and hence avoiding the generation of a long list of news of theatre events for Bob
(Herder, 2006, p:30).
In situations where the user interface or learning type is going to be adapted, it is
sufficient to use a stereotype model. However, when individual adaptations are required or
guidance specific to individuals or suggestion systems will be active, stereotype model might
fall short (Koch, 2000, p: 51).
Overlay model
Overlay models fundamentally regard student information as a subset of expert
information (Figure 4). Overlay models are based on the premise that the materials presented
by intelligent learning systems will ensure complete correlation between the student’s
knowledge and the expert’s knowledge. One of the downsides of this approach is its
disregard to the fact that students may have beliefs that may not be included in experts’
information (Beck, Stern and Haugsjaa, 1996).
Figure 4. Overlay Model (Beck, Stern, and Haugsjaa, 1996)
The success of the overlay model tightly depends on dividing the expert information
into as small elements as possible. In the overlay model, what the student knows and does not
know may be specifically shown. The information level the student has on each information
unit may be shown as “1-0”, or “knows-does not know”. More advanced overlay models may
indicate the knowledge status of the students by grades (an integer or probability measure)
441
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
(Brusilovsky, 1994). The decision that the system makes on the knowledge level of the
student in relation to each knowledge unit is the basis of the overlay model. The use of
probability techniques is generally preferred in making this decision. Since it is the approach
that deals with uncertainty best, Bayesian Networks are the most widely used probability
technique (Özmert
Perturbation model
The overlay model has a disadvantage in that it does not contain the mistakes and
misconceptions of students. However, students have misconceptions about subject fields in
general. For that reason, overlay models have been expanded to allow representation of the
false information students might have (Figure 5), and perturbation model has been
constructed (Beck, Stern and Haugsjaa, 1996). The perturbation model differs the overlay
model since it doesn’t perceive students knowledge as a simplification of expert knowledge,
but rather like perturbations over the expert knowledge (Webber, 2004, s.711). The
perturbation model may contain users’ errors or bugs and the reasons why users encounter
these errors. The adaptive system can present learning material, concepts, subjects or topics
with which users don’t know by obtaining information from the perturbation model. Adaptive
systems can also provide users with explanations, annotations to learn accurately, or assist
users in correcting errors (Nguyen and Do, 2008). In other words, perturbation model make it
possible to better evaluate student errors and transform students’ false information
pedagogically meaningful data (Beck, Stern and Haugsjaa, 1996).
Figure 5. Perturbation Model (Beck, Stern and Haugsjaa, 1996)
Fuzzy logic
At the center of using overlay model for student modeling lies the idea of encoding
the information of the expert relevant to the situation and making decisions using this
information. In first student models, expert information was stated by dual codes based on
442
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
the if-then structure. For instance, let us consider the statement “The room will be dark unless
the lights are on.” The interpretation of a statement as absolute right or wrong in this way is
known as classic logic. However, the daylight coming in through the window may also have
an effect on the darkness of the room. As seen in this example, in some cases, more than one
variable may have an effect on a consequence. The experts’ need to support their
explanations on certain information (if-then) with “else” revealed the inadequacy of the 0-1
coding. However, the fact that the strength of the light through the window determines the
darkness level of a room will be proved the “else” coding inadequate as well, and created the
need for a fuzzy logic. Fuzzy logic is based on the acceptance that a value of any situation
can be any continuous value between 0 and 1 (Baykal and Beyan, 2004; Jensen, 1997, Kadie,
Hovel and Horvitz, 2001).
Fuzzy logic may be defined as a strict mathematical order to communicate and work
with uncertainties. As is known, in statistics and probability theory, we work with certainties
as opposed to uncertainties, but the environment we live in is rather full of uncertainties.
Hence, in order to understand the deduction capability of humans, it is necessary to work
with uncertainties. The mathematics only allows extreme values, and that is the reason why it
is difficult to model complicated systems with classis mathematical methods, since the data
has to be complete. Fuzzy logic removes this obligation,allows for a more qualitative
definition (Constantin, 1996; Aktaran: Yavuz and Karaman, 2003), and provides a
mechanism to mimic human decision-making (Frias-Martinez, Magoulas, Chen, Macredie,
2005). Fuzzy logic defines a framework in which the inherent ambiguity of real information
can be captured, modeled and used to reason under uncertainty. Fuzzy Clustering, Fuzzy
Association Rules, etc. are the combination of fuzzy logic and machine learning techniques to
produce user models. It is also possible to capture user models with a machine learning
technique and use fuzzy logic inference to implement the personalization engine (FriasMartinez, Magoulas, Chen, Macredie, 2005).
Machine learning techniques
The field of machine learning, a sub category of the broader field of artificial
intelligence, deals with the design and development of algorithm and techniques that help
computers learn (Knight, 2004, p.4). Many successful applications from face recognition
systems (Ergezer, 2003) to risk assessment in medical interventions (Toprak et al., 2003)
443
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
have been developed in the field of machine learning, which seeks to answer the question of
how computer programs that automatically improve their own performance based on
experience and expand their database may be constructed (Mitchell, 1997). As information
and communication technologies become more integrated into daily life and the need to
develop systems that can solve the complicated problems encountered becomes more
apparent, machine learning has become more critical. Machine learning is predominantly
used in the data mining process where previously unknown but meaningful data is attempted
to be obtained from within extensive mass data (Gopal, 2000, p:18; Mitchell, 1997).
The focus of the researches made in the field of machine learning is extracting
meaningful information out of large data masses using numeric and statistical methods
(Vrakas and Vlahavas, 2008). The general goal of machine learning is to obtain the same
results in a more efficient, correct and less costly way by automating the steps people follow
in a decision-making process, using computers (Jordan, 2008). By directly observing user’s
interaction with the system and allowing the systems to make decisions based on these,
machine learning is quite functional in student modeling (Alessandro, 2006, p:27). Machine
learning enables to model the uncertainities in complex problems and to inference meaningful
results from the modeled information. Various techniques called as predictive statistical
models are developed to deal with these uncertainties in the field of machine learning.
Predictive statistical models enable the estimation of certain directions of human behavior
such as goals, actions, and preferences (Zukerman and Albrecht, 2001). Markov models,
artifical neural networks, and Bayesian networks are some of predictive statistical models.
Bayesian Networks
Probability, among the various ways suggested to model uncertainty, is a
mathematical structure that may act in harmony with certain rules (Nokelainen et al., 2001).
The probability theory is interested in events and their probability of happening. Bayesian
networks are derived from Bayesian probability theory that enables to compute the posterior
probabilities from uncertainties, prior probabilities, and casual relations (Niedermayer, 1998).
In classic statistics, the frequency value is required, which is derived from the (observable)
occurence number of a certain event divided by the total number of observations throughout a
long period. In Bayesian interpretation, however, the degree of subjective beliefs regarding
the probability of happening of an event is important (Nokelainen et al., 2001).
444
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Bayesian networks are directed non-cyclical graphs where variables are shown with
nodes, and conditional probabilities are shown with links (Jameson, 1998). Bayesian method
is one of the basic methods used to overcome uncertainty in student model systems (Jameson,
1996) and has been used in the student modeling process of many adaptive learning
,
dynamically updated data sets may be used and customized inference mechanisms for each
students may be created (Witten and Frank, 2000). While both fuzzy logic and bayesian
network can be used to handle uncertainty (represent subjective belief), fuzzy logic focus on
the degree of membership of an element in that set, bayesian network focus on degree of
belief of a certain person that a variable is in a set (Frias-Martinez, Magoulas, Chen,
Macredie, 2005; Heckerman and Shortcliffe, 1992).
Techniques used in constructing a student model, such as bayesian network, fuzzy
logic, perturbation model, overlay model, and stereotype model, may be used separately as
well as handled together in the student model. For instance, a student model based on the
misconceptions of students about “web design” subject can be constructed with fuzzy logic
and perturbation model duo.
Updating the student model
The student model is constructed prior to the student using the system and is updated
in the light of the data obtained in the learning process of the student (Shareef and Kinshuk,
2003). The methods used in updating the student model may be summarized under two titles;
performance measures and student activity analysis.
Performance measures
When a student answers a question, the answer is analyzed by the system and this
process is called performance measurement (Brusilovsky, 1994). For instance, in a practice
and drill software, depending on the student’s right or wrong answer related to a certain
learning objective, the variable indicating the knowledge status of that student is updated.
This way, decisions can be made if the system should pose any more questions on the same
learning objective or move on to questions concerning a new learning objective.
Student activity analysis
445
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
By analyzing the activities like the nodes clicked on, the duration spent on the nodes,
the selected features of the system like help, note taking, the queries made while a student
uses a system, the information like which subjects are studied or explanation types are
selected etc., can be updated.
In order to update a student model, both performance measures and the student
activity analysis may be used at the same time.
Student Model Trends
One of the most important features of student models is the fact that they are
accessible and usable by different applications. An valuable database may be created with the
characteristics brought together in a user model and the interpretation of these characteristics.
When a different application requires some of these characteristics of the user, recollecting
similar data leads to wasted effort, time, and money. For this reason, it is important that the
information in a student model can be reusable. When a reusable student model is
constructed, it should go with certain standards (Kay, 2000). The IMS Learner Information
Package Information Model (IMS LIP) portrays the structure of user data to allow
interoperability between different systems. The core structures of the IMS LIP are based
upon: accessibilities, activities, affiliations, competencies, goals, identifications, interests,
qualifications, certifications and licenses, relationship, security keys, and transcripts (IMS,
2009). The Learning Technology Standards Committee (LTSC) has defined specifications
like P1484 and P1484.2 for student data. While the purpose of P1484 is to specify common
architecture for CAI systems (LTSC, 2001), the purpose of P1484.2 is to describe portable
learner records (IEEE, 2000). P1484.2 (The Public and Private Information for Learners
Standard) is a data interchange specification and used for communication among cooperating
systems (IEEE, 2000). In addition to these, in order for the e-learning environment to offer
appropriate content for different students in accordance with the information in the student
model, the contents need to be designed as learning objects and metadata that defines each
learning object should be created. It is the only way to determine by the learning system
which contents are suitable for student characteristics (Kay, 2000).
Conclusion
Since the behaviors of an adaptive educational hypermedia largely depend on the
individual learning needs and preferences of students, student modeling is the key element for
446
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
personalization. This study is conducted to explain the major concepts/steps in student
modeling by focusing on the student model issue using a holistic approach. Therefore, the
concept of student model is defined, and gathering data related to the student characteristics,
model constructing, and updating phases in the student modeling process are addressed in
detail.
A student model is designed and developed in the direction of decisions on the various
aspects of the system, which will be adapted to the user. Student models intended for a
variety of purposes may be constructed, such as determining preferences of students in the
learning process, recognizing their plans and solutions, evaluating their performances and
problem-solving skills, etc. The degree of addressing real world complexities and the ability
of the techniques to handle missing/uncertain information about student and context indicate
the level of success of the student model. Another important feature of student models is their
accessibility by and availability to different learning applications.
In conclusion, in order to improve the performance of adaptive educational
hypermedia, the student model process should be carefully designed and constructed.
Therefore, it is of importance to analyze in detail all the elements of the student model
process and compare the efficiency of alternative methods/techniques for the same learning
conditions. Also, the case studies and experimental studies aiming to determine which
modeling techniques appropriate for different student/context characteristics and available
data will contribute to more value for implementation of adaptive educational hypermedia.
References
Alessandro, A. (2006). Inferring Dynamic Learner Behavior For User Modeling In
Continuously Adapting Hypermedia. Ph.D Thesis, University of Tennessee.
Alotaiby, F. T. (2005). A Component Based Functional Model For E-Learning Systems.
Ph.D Thesis, George Mason University.
Barra, M., Malandrino, D., Scarano, V. (2003). Common Web Paths in a Group Adaptive
System, HT’03, August 26–30, 2003, Nottingham, United Kingdom.
Baykal, N. & Beyan, T. (2004).
.
Beaumont, I. (1994). User Modeling and Hypertext Adaptation in the Tutoring System
ANATOM-Tutor. Proceedings of the Fourth International Conference on User
Modelling UM’94. August 15-19, Hyannis, Massachusetts, USA.
Beck, J., Stern, M. & Haugsjaa, E. (1996). Applications of AI in Education. ACM
Crossroads. Special issue on artificial intelligence, 3(1), 11-15.
447
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Bohnert, F. (2008). Constraint-Aware User Modelling and Personalisation in Physical
Environments. In Adjunct Proceedings of the Sixth International Conference on
Pervasive Computing (Pervasive-08), 167–172, Sydney, NSW, Australia, 2008.
Brusilovsky, P. (1994). The Construction and Application of Student Models in Intelligent
Tutoring Systems, Journal of Computer and System Sciences International, 32( 1),
70–89.
Brusilovsky, P. (1998). Methods and Techniques of Adaptive Hypermedia. Adaptive
Hypertext and Hypermedia. P. Brusilovsky, A. Kobsa and J. Vassileva (Editors), (p.
1-44). Boston: Kluwer Academic Publishers.
Butz, C.J., Hua, S. & Maguire, R.B. (2006). A Web-based Bayesian Intelligent Tutoring
System for Computer Programming, Web Intelligence and Agent Systems: An
International Journal, 4(1). 77-97.
De Bra, P. (1996). Teaching Hypertext and Hypermedia through the Web, Journal of
Universal Computer Science, 2(12), 797-804.
De Bra, P., (1998). Adaptive Hypermedia on the Web: Methods, techniques and applications,
Proceeedings of the AACE WebNet'98. 220-225, AACE, Orlando, Fl.
De Bra, P. (1999). Design Issues in Adaptive Hypermedia Application Development,
Proceedings of the Second Workshop on Adaptive Systems and User Modeling on the
World Wide Web, 29-39, Toronto and Banff, Canada,
Ebner, T., Magele, C. & Dietinger, T. (1999). Design and Implementation of Interactive,
Web-Based Courses, Proceedings of WebNet 99 World Conference on the WWW and
Internet, P., De Bra, J. Leggett (Editors), Charlottesville: AACE, 319-324.
Ergezer, H. (2003).
Yöntemleri. MS Thesis
University.
Frias-Martinez, E. Magoulas, G. D., Chen, S. Y., Macredie, R. D. (2005) Modeling Human
Behavior in User-Adaptive Systems: Recent Advances Using Soft Computing
Technique. Expert Systems with Applications. 29(2).
Fröschl, C. (2005). User Modeling and User Profiling in Adaptive E-learning Systems. MS
Thesis, Graz University of Technology.
Gopal, K. (2000). An Adaptive Planner Based On Learning of Planning Performance. MS
Thesis, Texas A&M University.
Heckerman, D. and Shortcliffe, E.H., From certainty factors to belief networks. Artificial
Intelligence in Medicine. 4(1). 35-52.
Herder, E. (2006). Forward, Back and Home Again Analyzing User Behavior on the Web.
Ph.D Thesis, University of Twente.
Holt, P., Dubs, S., Jones, M., & Greer, J. (1994). The State of Student Modeling. Student
Modelling. In The Key to Individualized Knowledge-Based Instruction (NATO ASI
Series). J. E. Greer and G. I. McCalla (Editors), (p. 3-35). Berlin: Springer-Verlag.
Höök, K. (1998). Evaluating the Utility and Usability of an Adaptive Hypermedia System.
Knowledge-Based Systems, 10(5), 311-319.
IMS. (2009). IMS Learner Information Package Specification. Retrieved 03/03/2009, from
http://www.imsglobal.org/profiles
448
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
IEEE. (2000). IEEE P1484.2. PAPI: Public and Private Information. Presentation to LTSC
Learner Model WG. Retrieved 03/03/2009, from http://ltsc.ieee.org/meeting/200003/
doc/papi--20000314--farance.ppt
LTSC. (2001). IEEE P1484.1/D9. Draft Standard for Learning Technology -Learning
Technology Systems Architecture. Retrieved 03/03/2009, from http://ltsc.ieee.org/
wg1/files/IEEE_1484_01_D09_LTSA.pdf
Jameson, A. (1996). Numerical Uncertainty Management in User and Student Modeling: An
Overview of Systems and Issues. User Modeling and User Adaptive Interactions. 5.
193- 251.
Jameson, A., Paris, C. & Tasso, C. (1997). ‘‘Reader’s Guide,’’ in User Modeling, Proc of the
Sixth International Conference UM97, Springer-Verlag, New York.
Jameson A. (1998). User Modeling: An Integrative Overview. Tutorial ABIS98:Workshop on
Adaptivitiy and User Modeling in Interactive Software Systems, FORWISS Report.
Jensen, F. V. (1997). Bayesian Networks and Influence Diagrams. Proceedings of The 1st
European Conference for Information Technology in Agriculture, 15-18 June The
Royal Veterinary and Agricultural University, Copenhagen, Danimarka, (p. 429-440).
Jordan, A. G. (2008). Frontiers of research and future directions in information and
communication technology. Technology in Society. 30(3-4). 388-396.
Kadie, C. M.. Hovel, D and Horvitz, E.. (2001). MSBNx: A Component-Centric Toolkit
for Modeling and Inference with Bayesian Networks. Microsoft Research
Technical Report
Kay, J. (1995). The UM Toolkit for Cooperative User Modeling. User Modeling and UserAdapted Interaction (UMUAI), 4(3). 149-196.
Kay, J. (2000). User Interfaces for All, chapter User Modeling for Adaptation, p.p. 271–294.
Human
Factors
Series.
Lawrence
Erlbaum
Associates,
Inc.,
http://www.cs.usyd.edu.au/~judy/Homec/Pubs/ch18.pdf
Kelly, D. (2005). On the Dynamic Multiple Intelligence Informed Personalization of the
Learning Environment. Ph.D Thesis, University of Dublin.
Knight, T. P. (2004). MARIA: A Multi-Layered Unsupervised Machine Learning Algorithm
Based On The Vertebrate Immune System. PhD Thesis, The University of Kent at
Canterbury.
Kobsa, A. (2001). Generic User Modeling Systems. User Modeling and User-Adapted
Instruction, 11(1-2), 49-63.
Koch, N. (2000). Software Engineering for Adaptive Hypermedia Systems: Reference Model,
Modeling Techniques and Development Process, Ph.D Thesis, Ludwig-MaximiliansUniversity of Munich.
Limongelli, C., Sciarrone, F., Temperini, M., Vaste, G. (2009). Adaptive Learning with the
LS-Plan System: A Field Evaluation. IEEE Transactions on Learning Technologies.
2(3). 203-215.
Mitchell, T. (1997). Machine Learning. U.S.A.: McGraw-Hill
449
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Niedermayer, D. (2008). An Introduction to Bayesian Networks and their Contemporary
Applications. Innovations in Bayesian Networks. D. E. Holmes and L. C. Jain
(Editors). Series: Studies in Computational Intelligence, Vol. 156, 117-130.
Nielsen, J. (1998), Personalization is over-rated, Retrieved 03/03/2009, from
http://www.useit.com/alertbox/981004.html.
Nguyen, L. & Do, P. (2008). Learner Model in Adaptive Learning. Proceedings of World
Academy of Science, Engineering and Technology, 35, 396-401.
Nokelainen, P., Silander, T., Tirri, H., Nevgi, A. & Tirri, K. (2001). Modeling Students’
Views on the Advantages of Web-Based Learning with Bayesian Networks.
Proceedings of The 10th International PEG2001 Conference, June 23-26 Tampere,
Finland.
WebYüksek Lisans Tezi,
Hacettepe Üniversitesi.
Rich, E. (1979). User Modeling Via Stereotypes, Cognitive Science, 3, 355-66.
Sas, C., Reilly, R. & O’Hare, G. (2003). A Connectionist Model of Spatial Knowledge
Acquisition in a Virtual Environment. Proceedings 2nd Workshop on Machine
Learning, Information Retrieval and User Modeling, 9th Int. Conf. on User
Modelling, 40-48.
Shareef, A. F. & Kinshuk, D. (2003). Student Model for Distance Education System in
Maldives. In Proceedings of E-Learn 2003 A. Rossett (Editor), (p. 2435-2438).
Arizona, USA: AACE.
Shute, V. J. & Psotka, J. (1996). Intelligent Tutoring Systems: Past, Present and Future. In
Handbook of Research on Educational Communications and Technology. D. Jonassen
(Editor). (p. 570-600). New York: Macmillan.
Somyürek, S. (2008).
Surjono, H. D. (2007). The design and implementation of an adaptive e-learning system, The
International Symposium Open, Distance, and E-learning (ISODEL 2007), Denpasar,
Indonesia, 13-15 November 2007.
Surjono, H. D. & Maltby, J. R. (2003). Adaptive Educational Hypermedia Based on Multiple
Student Characteristics. Advances in Web-Based Learning - ICWL 2003. Springer
Berlin / Heidelberg. Volume 2783/2003, 442-449.
Stern, M. K. (2001). Using Adaptive Hypermedia and Machine Learning to Create
Intelligent Web-Based Courses. Ph.D Thesis, University of Massachusetts.
& Arslan, A. (2003). Genetik Algoritmalarla Makina
2003. 3Çukurova Üniversitesi ,Adana.
Triantafyllou E., Demetriadis S., Pomportsis A. (2003). Adaptive Hypermedia and Cognitive
Styles: Can Performance Be Influenced?. In Proceedings of the 9th Panhellenic
Conference in Informatics, Thessaloniki.
450
Somyürek, S. (2009). Student modeling: Recognizing the individual needs of users in e-learning environments.
International Journal of Human Sciences [Online]. 6:2. Available: http://www.insanbilimleri.com/en
Trousse, B. (2000). Evaluation of the Prediction Capability of a User Behaviour Mining
Approach for Adaptive Web Sites, In Proceedings of the 6th Computer-Assisted
Information Retrieval Conference (RIAO), France.
Tsiriga, V. & Virvou, M. (2003).Initializing Student Models in Web-Based ITSs: A Generic
Approach. Proceedings of the 3rd IEEE International Conference on Advanced
Learning Technologies (ICALT 2003),(p. 42–46), 9-11 July, Athens, Greece.
Vrakas, D. & Vlahavas, I. P. (2008). Artificial Intelligence for Advanced Problem Solving
Techniques. U.S.A.:Information Science Reference.
Webber, C. (2004). From Errors to Conceptions. An Approach to Student Diagnosis. J. C.
Lester, R. M. Vicari, & F. Paraguacu, (Eds.), Intelligent Tutoring Systems: 7th
International Conference, 710-719 Berlin: Springer-Verlag Berlin & Heidelberg
GmbH & Co. K.
Wenger, E. (1987). Artificial Intelligence and Tutoring Systems: Computational and
Cognitive Approaches to the Communication of Knowledge. Los Altos, CA: Morgan
Kaufmann Publishers, Inc.
White, R. W., Bailey, P. and Chen, L. (2009). Predicting User Interests from Contextual
Information, SIGIR’09, July 19–23, 2009, Boston, Massachusetts, USA.
Data Mining: Practical Machine Learning Tools and
Techniques with Java Implementations. San Diego, CA: Morgan Kaufmann
Publishers, Inc.
e Makinesi.
. 3Üniversitesi, Adana.
Zhang, J. & Ghorbani, A. A. (2007). GUMSAWS: A Generic User Modeling Server for
Adaptive Web Systems. Fifth Annual Conference on Communication Networks and
Services Research (CNSR 2007), 117-124, 14-17 May 2006, Fredericton, New
Brunswick, Canada, IEEE Computer Society
Zhou, Y. & Evens, M.W. (1999). A practical student model in an intelligent tutoring system.
Proceedings 11th IEEE International Conference on Tools with Artificial Intelligence,
Evanston, Illinois, 13–18.
Zukerman, I. & Albrecht, D. W. (2001). Predictive Statistical Models for User Modeling.
User-Adapted Interaction. 11, (1-2), 5-18.