Students' questions are useful for their learning and for teachers' pedagogical adaptation. However, the volume of questions asked online by students may prevent teachers from dealing with each question (e.g. MOOC or large university cohort).
We address this issue mainly in the context of a hybrid training program in which students ask questions online each week, using a flipped classroom approach, to help teachers prepare their on-site Q&A session. Our objective is to support the teacher to determine the types of questions asked by different groups of learners. To conduct this work, we developed a question coding scheme guided by student’s intention and teacher’s pedagogical reaction. Several automatic classification tools have been designed, evaluated and combined to categorize the questions. We have shown how a clustering-based model built on data from previous sessions can be used to predict students' online profiles using exclusively the nature of the questions they ask. These results allowed us to propose three alternative questions’ organizations to teachers (based on questions’ categories and learners’ profiles), opening up perspectives for different pedagogical approaches during Q&A sessions. We have tested and demonstrated the possibility of adapting our coding scheme and associated tools to the very different context of a MOOC, which suggests a form of genericity in our approach.