Tailored scaffolding for meta-cognitive skills during analogical problem solving

Although examples play a key role in cognitive skill acquisition, research demonstrates that learning outcomes are heavily influenced by the meta-cognitive skills students bring to bear while using examples. This dissertation involves the design, implementation and evaluation of the Example Analogy (EA)-Coach, an Intelligent Tutoring System that provides adaptive support for meta-cognitive skills during a specific type of example-based learning known as analogical problem solving (APS, using examples to aid problem solving). To encourage the targeted meta-cognitive skills, the EA-Coach provides multiple levels of scaffolding, including an innovate example-selection mechanism that aims to choose examples with the best potential to trigger learning and enable problem solving for a given student. To find such examples, a key factor that needs to be taken into account is problem/example similarity, because it impacts the APS process. However, full understanding has yet to be reached on how various levels of similarity between a problem and an example influence students' APS behaviours and subsequent outcomes. Here, we provide a novel classification of problem/example differences and hypotheses regarding their impact on APS. In particular, we propose that certain differences between a problem and an example may actually be beneficial in helping students learn from APS, because they promote the necessary meta-cognitive skills. However, given the great variance in terms of knowledge and meta-cognitive skills that exists between students, a key challenge with our approach is how to select examples that provide enough scaffolding for different learners. Our solution to this challenge involves a two-step decision-theoretic process. First, the EA-Coach student model, which corresponds to a dynamic Bayesian network, is used to predict how a candidate example will help a student solve the problem and learn from doing so. Second, the model's prediction is quantified via a utility function, which assigns an expected utility to the candidate example. This process allows the framework to present to the student the example with the highest expected utility for enabling learning and problem solving. We evaluated this approach via a controlled laboratory study, which demonstrated the EA-Coach's pedagogical effectiveness for supporting problem solving and triggering meta-cognitive skills needed for learning during APS.Science, Faculty ofComputer Science, Department ofGraduat