Strategies for problem-solving : a model to classify problem-solving behaviours

décembre 2024

Directeur(s) de thèse : Thierry Vieville

Résumé de thèse

Learning regulation is the set of skills that facilitate learning. This includes motivational and emotional skills, as well as meta-cognition skills, such as monitoring goals and strategies. Learning situations are present in every-day life, and mostly take the form of open problems, problems for which there is an unknown solution or method to achieve them. Open problem solving requires learning unknown elements, which requires regulation skills. Understanding the strategies used by learners to solve open-ended problems allows us to better understand the regulation processes and therefore the learning process. Regulatory processes can be influenced and trained, and this understanding of the processes could lead, to better-suited teaching methods. This thesis presents a model representing the learning process of a learner confronted with an open problem, as well as a method allowing, based on external observations of actions carried out by learners, to group problem-solving sequences depending on the strategies used. The analysis methods use clustering algorithms, which using dynamic time warping as a measure of similarity between problem-solving sequences. This dynamic time warping measure is applied to symbolic data, which can be compared through the use of proper metrics. In the context of small datasets, a metaclustering method is presented for robustness purposes. Two case studies use the method defined here to show the potential uses of such a method but also to discuss its limitations. The first, CreaCube, is the study of a creative problem-solving task, in which we try to understand the processes of creativity. Second, Outer Wilds tries to extend the analysis to long problems.