Affective modelling is receiving increasing attention due to its recognized relevance in learning. It is considered a particularly challenging task for two main reasons. First, ground truth is unattainable, and thus it invariably requires indirect measures and approximations. Second, experimental data are limited in quantity and quality due to high costs and significant noise levels. In our previous work (Baschera et al., 2011), we have developed an engagement dynamics model in spelling learning that can adapt the training to individual students based on data-driven identification of engagement states from student input.
In this study, we explore the defnition of a general framework for modelling engagement dynamics in human learning. In particular, we focus on developmental dyslexia and dyscalculia. We argue that the assumption of similar engagement patterns in the two cases is justified and, thus, that a similar engagement model would be beneficial. We provide a detailed assessment of similarities and dissimilarities of the two cases of developmental dyslexia and dyscalculia in terms of learning domain, student model, and available data. Furthermore, we analyse the reusability of the engagement model for spelling learning and define desirable properties of a general model of engagement dynamics for software tutoring.
Our engagement model for spelling learning can adapt the training to individual students based on data-driven identification of engagement states from student input. Engagement states are modelled using a dynamic Bayesian net representation, which jointly represents the influences of Focused and Receptive states on learning, as well as the decay of spelling knowledge due to Forgetting. The presented causal model can be investigated and exhibits coherent conclusions.
Although developmental dyscalculia and dyslexia show comorbidities and similiarites in engagement states, our comparison of the two learning disabilities has revealed that there are significant differences in the two learning domains, the modelling of the student in the respective computer-based training environment, as well as in the availability and interpretation of the experimental data. Especially the indicator function (indicating whether the child is attentive) and the features used in the model for spelling learning (Baschera et al., 2011) are specific to the learning domain. Therefore, we suggest a comprehensive feature set that can be used for different learning domains and environments. Features of this set are particularly suitable for hierarchical learning domains such as mathematics learning.