xCALE

With the advent of MOOCs, and the success of exercise-based platforms, large-scale online environments are becoming widespread, both in distant learning and in blended learning. Self-Regulated Learning (SRL) is known to have a good potential on autonomy development and on maintaining motivation for learners, both in MOOCs , and in blended learning within exercise-based platforms. Developing Self-Regulated Learning Strategies is also known to have a positive impact on academic achievement. Our main objective in this project is to investigate how to support successfully self-regulated learning at a large scale, with an approach that aims to estimate acquired skills levels and metacognitive levels about students to provide appropriate interventions. This approach pursues the objective of fostering students’ autonomy through Open Learner Models (OLM). Along the same lines as Conati et al. (2018), who demonstrates that effective support assumes that models are interpretable by users, we defend that support should be able to provide some forms of explanation on learning process to stimulate reflection on metacognition, beyond simple activity recommendations. We are interested in Bayesian modelling techniques to acquire and update learners’ models as OLMs (about their cognitive and metacognitive skills and progression), with an application on data corpus resulting from available large-scale learning platforms comprising MOOCs series (IMT Altantique courses available on FUN and Edx platforms) and courses from a web platform dedicated to learning programming with practical exercises in Secondary Education (France IOI). We expect from this project a good potential for generalizability and transferability, since we dispose of real life data. In concrete terms, xCALE project proposes to develop, experiment and evaluate a generic approach that allows to provide a personalized and interpretable support for (i) skills acquisition and (ii) self-regulated learning to support learners’ autonomy. Personalized learning and self-regulated learning will be based on learners’ models - having features for personalization and for self-regulation - and Bayesian modelling techniques. Expected results and verifiable indicators include what follows: - A methodology to develop and evaluate Open Learner Models on learners’ skill levels in didactically well-defined disciplines that provide personalized and interpretable insights to users. Proven examples of personalized interpretable interventions in programming and algorithmic courses, based on learner data and expert knowledge practitioners will be provided. - A generic model on metacognitive processes, based on learners’ progression and SRL-based learning interactions. This model will predict metacognitive strategies to provide relevant interventions. - A recommendation engine that will provide, on the one hand, personalized activities for skills acquisition according to SRL support, and on the other hand, relevant and personalized interventions to develop SRL (including visual dashboards, dedicated tools, such as time planning or goals management, guidance, etc.). - An impact study on the transformation of teaching practices, as well as the dissemination of SRL support proposals among other courses and disciplines.