Advances in Human-Computer Interaction / 2014 / Article / Tab 2

Research Article

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.

TargetClassesDBNSBNNBDTSVM

Interaction experience trendFlow, stuck, and off-task 75.6369.3163.7164.0769.09
StressNo (calm), low, moderate, and high61.0947.01N/AN/AN/A
ConfusionNo (confidence), low, moderate, and high60.0253.71N/AN/AN/A
BoredomNo (interest), low, moderate, and high79.9563.45N/AN/AN/A
FrustrationNo (satisfaction), low, moderate, and high67.4655.36N/AN/AN/A

Interaction experience trendPositive, negative82.2573.1268.7869.2672.23
StressCalm to low stress, moderate to high stress82.1868.95N/AN/AN/A
ConfusionConfidence to low confusion, moderate to high confusion81.8867.41N/AN/AN/A
BoredomInterest to low boredom, moderate to high boredom90.9771.04N/AN/AN/A
FrustrationSatisfaction to low frustration, moderate to high frustration85.3869.02N/AN/AN/A