| dc.description.abstract | The research presented in this thesis aimed at non-intrusively detecting a learner s emotional state in a programming tutorial environment. These results could be used to adapt instructions and feedback to students in an online programming tutorial, potentially leading to more effective and motivating learning.
Detection of a learner s emotional state was done by collecting and analysing 23 participants keystroke dynamics (how people type on their keyboard), and additionally pulse (heart rate) for five participants, in an online JavaScript tutorial developed for this research. Participants self-reported their emotional state, selecting one of six predefined states, hypothesised to be relevant for a learning situation. These emotions were: Bored, concentrated, confused, delighted, frustrated and surprised.
Both multiclass and binary classifiers were trained and tested on the dataset. In the binary classifiers, five classes were aggregated and classified against the sixth. Classification was tested on the whole population, and on individual participants. Every experiment was done with and without pulse features included to see if pulse influenced the classification.
Binary classifiers, using the whole population as a dataset, yielded the most promising results with accuracies ranging between 60\% and 100\%. Pulse was not found to give a better classification in this research. No conclusive results may be given however, as there are limitations in both the dataset and how the pulse feature was implemented. Still, this research does show promising results for non-intrusive emotion detecting in a programming environment. | |
