A Hierarchical Probabilistic Framework for Recognizing Learners’ Interaction Experience Trends and Emotions
Table 2
Model inference accuracy. Outright classification is done by assigning each instance to the class with the highest probability (maximum a posteriori procedure). Participants’ matching self-reports are used as a ground truth. For the nonhierarchical approaches (NB, DT, and SVM), the inference is achieved only for the experienced trends.
Target
Classes
DBN
SBN
NB
DT
SVM
Interaction experience trend
Flow, stuck, and off-task
75.63
69.31
63.71
64.07
69.09
Stress
No (calm), low, moderate, and high
61.09
47.01
N/A
N/A
N/A
Confusion
No (confidence), low, moderate, and high
60.02
53.71
N/A
N/A
N/A
Boredom
No (interest), low, moderate, and high
79.95
63.45
N/A
N/A
N/A
Frustration
No (satisfaction), low, moderate, and high
67.46
55.36
N/A
N/A
N/A
Interaction experience trend
Positive, negative
82.25
73.12
68.78
69.26
72.23
Stress
Calm to low stress, moderate to high stress
82.18
68.95
N/A
N/A
N/A
Confusion
Confidence to low confusion, moderate to high confusion
81.88
67.41
N/A
N/A
N/A
Boredom
Interest to low boredom, moderate to high boredom
90.97
71.04
N/A
N/A
N/A
Frustration
Satisfaction to low frustration, moderate to high frustration