Information Technology Reference
In-Depth Information
the relevance degrees among concepts. Learner
testing records are also exploited. Fuzziness can
be used in domain knowledge representation. For
instance in (Cardinaels, Duval & Olivie, 2006)
confidence values for metadata facet values are
defined, in (Sicilia et al. 2005) the LOM and
SCORM standards are extended by defining im-
precise links (relationships) among learning items
and in (Goodkovsky, 2004) fuzzy structures are
used for domain knowledge representation. Fuzzy
clustering can be used to analyze accumulated
data involving learning activities and experiences.
For instance in (Dogan, Camurcu, 2009) learner
records are clustered to obtain pedagogically reli-
able information as feedback to tutors.
Constraint-based modeling
(Mitrovic &
Martin, 2007) is a representation scheme suited
for learner knowledge evaluation. A constraint
indirectly represents the solutions violating the
knowledge domain. The user's knowledge is
represented as a set of constraints that he/she
violates or not.
Genetic algorithms
employ evolution tech-
niques to find adequate solutions to problems
(Michalewicz, 1999). Candidate solutions to a
problem are represented as strings called chro-
mosomes. Crossover and mutation operators are
applied to existing candidate solutions in order to
produce new candidate solutions. A function (i.e.
fitness function) producing a numerical value is
used to evaluate a candidate solution's ability to
solve a problem. If this numerical value is below
a threshold, the corresponding candidate solution
is not retained in the pool of candidate solutions.
Genetic algorithms can be applied to perform
pedagogical tasks such as course planning and
learning content selection. For instance, in (Chen,
2008) a genetic algorithm approach is employed
to provide learning path guidance to learners
based on evaluated learner knowledge regarding
domain knowledge concepts. In addition, genetic
algorithms can be used in tasks such as optimiza-
tion of IES modules and contents (Koutsojannis
et al., 2007) and learner record analysis.
Reinforcement learning
is a machine learning
approach in which the task is to learn a policy for
choosing optimal actions to achieve certain goals
(Mitchell, 1997). Delayed reward is provided
for performed actions. So, the system is trained
to choose sequences of actions maximizing cu-
mulative reward. Not many IESs have employed
reinforcement learning. Reinforcement learning
can be applied to IES pedagogical tasks since it
resembles tutoring based on trial-and-error. The
pedagogical module applies instructional strate-
gies tailored to learner needs based on previous
interaction of the specific learner or learners with
similar characteristics. It assists in avoiding the
effort/cost in acquiring extensive pedagogical
knowledge. However, data involving simulated
learners are necessary in order to provide the
IES with an accurate initial pedagogical policy
(Iglesias et al. 2009).
Recently,
hybrid KR&R techniques
have started
to be used in IESs. Hybrid KR&R techniques com-
bine more than one KR&R technique (Medsker,
1995) and offer advantages in developing IESs
(Hatzilygeroudis & Prentzas, 2006). The main
goal of hybrid approaches is to surpass disadvan-
tages/limitations of each combined method and
simultaneously benefit from advantages of each
method. Various types of hybrid KR&R techniques
have been developed. According to the employed
combination model, the combined components
in the hybrid KR&R scheme may be distinct or
indistinguishable. Popular such hybrid KR&R
approaches involve combinations of RBR with
neural networks, combinations of neural networks
with fuzzy logic (i.e. neuro-fuzzy approaches),
combinations of genetic algorithms with other
approaches and combinations of CBR with other
approaches (e.g. RBR). For example, formalisms
integrating rules and neural networks are used in
representing human reasoning in the pedagogical
module (Hatzilygeroudis & Prentzas, 2004). In
(Stathacopoulou et al., 2007) a neuro-fuzzy ap-
proach is employed to evaluate learners' learning
style through classification. In (Papanikolaou et
Search WWH ::
Custom Search