Multi-relational Matrix Factorization Approach for Educational Items Clustering

Abstract

Pedagogical models development requires several steps, one of which is the mapping of tasks and skills, also known as the educational items clustering. This activity of clustering educational items usually requires the participation of domain experts. However, discovering the exact skills involved in performing the tasks is a complex activity for them. This paper aims at solving the task and skill-mapping problem by proposing an approach based on the Weighted Multi-Relational Matrix Factoring technique to help experts in this task. This approach relies on two types of relationship, the “ student does task” relationship and the “student has skills” relationship through a latent factor model to reconstruct the “ task requires skill” relationship, the latter being the mapping between tasks and skills. An evaluation conducted on a group of two hundred (200) students in lower 6th class in a general secondary school (Côte d'Ivoire), showed that this approach brought an improvement rate of about 82.8% of the skill-task mapping proposed by the experts in the field. This result confirms that our approach not only allows us to map tasks and skills but also to significantly improve the updating of curricula.