discussion forums are spreadly employed as learning tools in online courses, particularly in the Massive open online course (MOOC). Learners share opinions, express needs, and seek tutoring, and participate in discussions in the online forum. However, learner’s workstation generates massive information due to the number of MOOC participants, making it difficult to identify relevant information that can help and answer questions during the MOOC. Identifying and extracting knowledge from a MOOC discussion forum requires learner’s engagement in a collaborative and informative learning environment that enables knowledge exchange and information sharing. In this article we offer a new approach to explore forums, interactions and collaboration of learners online, in a knowledge building process, by an extraction framework and presentation of knowledge based on the characteristics of the text written in the learners' messages during the training. Our proposal consists in combining the pretreatment of the natural language by the TF-IDF metric, and the embedding of the words by Word2Vec, and then we will use the machine learning algorithm SVM for a semantic classification according to the analysis interactions model. Thus, we will apply the transformations and pretreatments on the messages posted in the forums by the participants in the MOOC, then the Word2Vec to represent each word as a vector, which will be concatenated to the features of the context TF-IDF. These vectors will form the data input of our Learning SVM machine algorithm, which aims to establish semantic relationships between concepts. The knowledge is then expressed as ontology for a representation of knowledge and an enrichment of our model.