Game Development Reference
In-Depth Information
Table 1. The Control-value theory of achievement emotions (Pekrun et al., 2007, p. 20)
Object focus
Appraisals
Value
Control
Emotion
Outcome/ Prospective
Positive (Success)
High
Medium
Low
Anticipatory Joy
Hope
Hopelessness
Negative (Failure)
High
Medium
Low
Anticipatory Relief
Anxiety
Hopelessness
Outcome/ Retrospective
Activity
Positive (Success)
Negative (Failure)
Positive
Negative
Positive/Negative
None
Irrelevant
Self
Other
Irrelevant
Self
Other
High
High
Low
High/Low
Joy
Pride
Gratitude
Sadness
Shame
Anger
Enjoyment
Anger
Frustration
Boredom
inference complexity is reduced, since parts of
the model can be used with instances of a larger
model. The domain is divided into classes, X 1 , X 2 ,
…, X n , with each class comprised of attributes A ij
є A(X i ) and each attribute corresponds to one or
more values V(A ij ) of a specific domain (Sucar
& Noguez, 2008). The dependencies are defined
at class level using a relational structure, where
attributes of a class can depend on attributes of
other classes. The conditional probabilities of
PRMs are defined in the same way as the con-
ditional probabilities of BNs, i.e. defining the
Markov Blanket for each attribute, e.g. parents
and children.
prospective outcome, activity and retrospective
outcome emotions. The DBNs were built using
observable variables that are related to the learner's
interaction behavior. Control and value apprais-
als are inferred from the state of these random
variables, which were identified from the work by
Del Soldato & Du Boulay (1995), McQuiggan et
al. (2008) and Pekrun et al. (2005). Initially this
work focuses only on motivational and cognitive
variables, since from the work of McQuiggan et
al. (2008), it was noted that physiological vari-
ables may result in improved model accuracy. In
addition, the existence of an emotion cannot be
ensured by using as direct evidence physiological
patterns (Pekrun, 2005).
The DBN corresponding to outcome-prospec-
tive emotions, hope, anticipatory joy, anticipa-
tory relief, anxiety and hopelessness, is shown in
Figure 2. Control and value appraisals are inferred
from variables related to the learner's motivation,
e.g. confidence and effort, and from variables
related to the learner's cognition, e.g. the latest
level of performance and level of difficulty. Con-
trol t-1 and Value t-1 represent the probabilities that
are transferred over time from the DBN corre-
sponding to outcome-retrospective emotions,
Building the Affective Student Model
The PRM structure corresponding to the learner's
emotions is shown in Figure 1. This model was
based on the control-value theory of 'achievement
emotions'. The dashed lines describe relations
between classes and the arrows represent condi-
tional probabilistic dependencies. Three Dynamic
Bayesian Networks (DBNs) were derived from
this PRM. Each DBN corresponds to one of the
emotion types defined by Pekrun et al. (2007), i.e.
Search Nedrilad ::




Custom Search