Multi-class classification aims at assigning each sample to one category chosen among a set of different options. In this paper, we present our work for the development of a novel system for multi-class classification of e-learning videos based on the covered educational subjects. The audio transcripts and the text depicted into visual frames are extracted and analyzed by Cognitive Computing tools, going over the traditional term-based similarity approaches. Preliminary experiments demonstrate effectiveness and capabilities of the system, suggesting that semantic analysis improves the performance of multi-class classification.