Cross-Domain Performance of Automatic Tutor Modeling Algorithms

In our recent work, we have proposed the use of multiple solution demonstrations of a learning task to automatically generate a tutor model. We have developed a number of algorithms for this automation. This paper describes the application of these domainindependent algorithms to three datasets from different learning domains (Mathematics, Physics, French). Besides verifying the applicability of our approach across domains, we report several domain specific performance characteristics of these algorithms which can be used to choose appropriate algorithms in a principled manner. While the Heuristic Alignment based algorithm (Algorithm 2) may be the default choice for automatic tutor modeling, our empirical finding suggest that the Path Pruning based algorithm (Algorithm 4) may be favored for language learning domains.