Subventions et des contributions :

Subvention ou bourse octroyée s'appliquant à plus d'un exercice financier. (2017-2018 à 2022-2023)

In the field of Intelligent Tutoring Systems, a critical modeling issue is to determine how the domain content relates to skills that we aim the student to learn. When this content represents tasks that can be succeeded or failed (question, exercise, etc.), the mapping of tasks to skills is the key to deliver a student skills assessment. When the content represents learning material, a tutoring system can use the mapping of skills to learning material in order to guide the student through the appropriate topics and reach specific learning goals.

A first objective of this research program is to design algorithms that can help map task to the required skills. We have made great progress in developing algorithms that can infer this mapping from data, and that refine and validate expert given mappings, but there is still room from improvement and we propose a number of research avenues in that respect.

A second objective of the program is to work on the mapping of learning content to skills. While the algorithms that map tasks to skill can use student performance data, such as test outcome, to infer the skills behind tasks, this type of data is not readily available for learning material. Material such as Wikipedia pages, textbooks chapters, or documents in general, is most often indexed by topics, not by the skills they can help acquire. Yet, learning objectives is best defined by skills, and the alignment of skills to concepts is not always clear, or even the one we would expect.

Therefore, standard techniques based on semantic distance between content and topics is deemed insufficient to determine the specific skills addressed by a given content. They cannot readily tell if the content is too difficult or already known by a student with a specific skill profile. Nor can they determined prerequesite skills and concepts that are involved in the content.

We plan to develop data driven approaches to determine the mapping of learning content to skills. These approaches will rely on textual features extraction techniques determine the related skills, and also make use of the techniques developed for tasks to skills mapping.