620
P.-Y. S. Hsueh et al.
or for detecting errors, etc.” To represent the domain models, both glass box model
and the black-box model have been developed. In the glass box model, domain
knowledge is represented by an expert system, which claims to be capable of
representing the experts’ problem-solving strategy and is accessible by the students.
The black-box model, on the contrary, does not claim this [83].
Second, the student model represents a variety of properties of the students. The
functionality of the ITS depends directly on the amount of information provided
by the student model; therefore, the student model should be as comprehensive as
possible. This includes all the factors that influence the student’s learning behavior,
e.g., the student’s cognitive and affective state or learning progress. Based on the
data collected, the student model infers new knowledge about the target student
and to leverage such knowledge for selecting a pedagogical strategy. Sometimes
it is called diagnostic model as it is used to diagnose the learning process. One
common way to model the student’s knowledge is to model it as a subset of the
expert knowledge base or as the difference from the expert knowledge base.
Third, the tutoring model unites the knowledge from both the domain and the
student model to adapt the tutoring strategies. It simulates the pedagogic behavior
of a human teacher to select content and plan for interventions.
Finally, the interactions with the learner are realized by the user interface model.
Because the presentation type often affects student’s learning capabilities, ITS
would need to account for different visualization strategies [82].
For the implementation of these models in the smart learning framework, various
artificial intelligence approaches have been deployed. In particular, two main
categories of methods emerge: symbolic AI and soft computing. The first category
of symbolic AI methods is knowledge-driven and requires the explicit representation
of knowledge bases, which means to explicitly express expert knowledge in
formal forms using a representation language. The expatiating process incorporates
significant cognitive work and therefore often results in high-quality knowledge
bases.
Contrarily, the second category of soft computing methods is data driven and
learns a desired behavior by investigating existent data. During the learning process,
the parameters of a mathematical model are adapted according to the observation of
learner performance. Generally, the learned internal representation is not directly
accessible and not understandable by humans, as most soft computing models
behave like black boxes.
Since both categories of method have their strengths and weaknesses, researchers
have investigated into the possibility of combining methods from both categories
into a knowledge symbiosis learning framework. (For comprehensive analysis,
please refer to [84].)
For example, [85] developed an ITS that aims to leverage the everyday working
knowledge of control center operators for incident analysis and diagnosis as
well as for service restoration. The system consists of two parts, DiagTutor and
CoopTutor. The first part is used for the training of fault diagnosis, and the second
part is used for restoration training. During diagnosis, control center operators
must analyze a combination of alarm messages and additional information, for
P.-Y. S. Hsueh et al.
or for detecting errors, etc.” To represent the domain models, both glass box model
and the black-box model have been developed. In the glass box model, domain
knowledge is represented by an expert system, which claims to be capable of
representing the experts’ problem-solving strategy and is accessible by the students.
The black-box model, on the contrary, does not claim this [83].
Second, the student model represents a variety of properties of the students. The
functionality of the ITS depends directly on the amount of information provided
by the student model; therefore, the student model should be as comprehensive as
possible. This includes all the factors that influence the student’s learning behavior,
e.g., the student’s cognitive and affective state or learning progress. Based on the
data collected, the student model infers new knowledge about the target student
and to leverage such knowledge for selecting a pedagogical strategy. Sometimes
it is called diagnostic model as it is used to diagnose the learning process. One
common way to model the student’s knowledge is to model it as a subset of the
expert knowledge base or as the difference from the expert knowledge base.
Third, the tutoring model unites the knowledge from both the domain and the
student model to adapt the tutoring strategies. It simulates the pedagogic behavior
of a human teacher to select content and plan for interventions.
Finally, the interactions with the learner are realized by the user interface model.
Because the presentation type often affects student’s learning capabilities, ITS
would need to account for different visualization strategies [82].
For the implementation of these models in the smart learning framework, various
artificial intelligence approaches have been deployed. In particular, two main
categories of methods emerge: symbolic AI and soft computing. The first category
of symbolic AI methods is knowledge-driven and requires the explicit representation
of knowledge bases, which means to explicitly express expert knowledge in
formal forms using a representation language. The expatiating process incorporates
significant cognitive work and therefore often results in high-quality knowledge
bases.
Contrarily, the second category of soft computing methods is data driven and
learns a desired behavior by investigating existent data. During the learning process,
the parameters of a mathematical model are adapted according to the observation of
learner performance. Generally, the learned internal representation is not directly
accessible and not understandable by humans, as most soft computing models
behave like black boxes.
Since both categories of method have their strengths and weaknesses, researchers
have investigated into the possibility of combining methods from both categories
into a knowledge symbiosis learning framework. (For comprehensive analysis,
please refer to [84].)
For example, [85] developed an ITS that aims to leverage the everyday working
knowledge of control center operators for incident analysis and diagnosis as
well as for service restoration. The system consists of two parts, DiagTutor and
CoopTutor. The first part is used for the training of fault diagnosis, and the second
part is used for restoration training. During diagnosis, control center operators
must analyze a combination of alarm messages and additional information, for