The automatic identification of core concepts addressed by a learning resource is an important task in favor of organizing content for educational purposes and for the next generation of learner support systems. We present a set of strategies for core concept identification on the basis of a semantic representation built using the open and available knowledge in the so-called Knowledge Graphs (KGs). Different unsupervised weighting strategies, as well as a supervised method that operates on the semantic representation, were implemented for core concept identification. In order to test the effectiveness of the proposed strategies, a human-expert annotated dataset of 96 learning resources extracted from MOOCs was built. In our experiments, we show the capacity of the semantic representation for the core-concept identification task as well as the superiority of the supervised method.