A three-step rectangular latent Markov modeling for advising students in self-learning platforms

Recent years have seen a growing interest in using technology to provide adaptive learning environments. In this vein, (self)learning environments that offer an automatic recommendation system play a fundamental role in supporting students’ learning with tailored feedback. In this aim, essential steps consist in collecting students’ responses and diagnosing their learning state throughout the learning process. This contribution proposes a three-step rectangular Latent Markov modeling to assess students’ abilities by analyzing sequences of response patterns to item-sets recorded at time intervals during the course. Each sequence corresponds to a measurement model that focuses on different topics. Furthermore, students’ ability is conceived as a multivariate latent variable that refers to diverse skills. The proposed approach consists of a three-step procedure: carrying out a multivariate Latent Class IRT model at each time point to find homogeneous groups of students according to their ability level; computing the time-specific classification error probabilities; fitting weighted logistic regressions to investigate the effect of socio-demographic and psychological variables on the initial and transition probabilities using the entries of the inverse of the classification error matrices as weights (BCH correction).