With the massive use of new learning technologies, such as mobile learning through on-line platforms, or more "traditional" live classes taught through virtual meeting platforms, the interaction between students and teachers can be poor or limited. The professor can monitor the students' reactions and appreciation with the use of polls, forms, and forums, but in general it is very difficult to understand how students feel, what they really like and how to give them an unforgettable learning experience, especially when they are many and in a heterogeneous environment. A method to automatically detect students' needs, based on unsupervised machine learning techniques is proposed. [For the complete proceedings, see ED639262.]