Overview
This paper, with Farnaz Tehranchi and Frank Ritter at Penn State, uses eye tracking to study how high- and low-performing learners behave differently while working through a cognitive tutoring system. Twelve learners took part. We asked whether performance can be predicted from visual attention alone, whether adding early performance improves the prediction, and whether age, gender, first language, where learners look, and the sequence of areas of interest (AOIs) matter.
Eye-movement variables were entered in mixed-effects logistic regression models that predict a correct answer on each question, and a classifier was trained on the eye-movement data.
Key points
- A model with eye-movement data alone predicted performance on each question with an area under the ROC curve of .76.
- Learners did better on the second set of questions, and the set variable was a reliable predictor.
- Among the non-eye-movement variables examined, only AOI coverage was significant: questions drawn from larger AOIs were answered correctly more often.
- High-performing learners paid more attention to the content that held answers to later questions.
- Design suggestions follow: make the content that contains answers larger and reconsider the order of content.
Event
42nd Annual Meeting of the Cognitive Science Society (CogSci 2020).
Citation
Tehranchi, F., Ritter, F. E., & Chae, C. (2020). Visual attention during e-learning: Eye-tracking shows that making salient areas more prominent helps learning in online tutors. In Proceedings of the 42nd Annual Meeting of the Cognitive Science Society (pp. 3164–3170).