Abstract: To solve the problem of false negative examples and zero loss, the transformer network structure is used to extract the entity description information in the knowledge base. The self attention mechanism with constraints is used to find the representation subspace that can best express the entity meaning to enhance the entity representation ability. The idea of counter generated network is introduced to generate negative samples,which improves the knowledge representation ability. In the construction of the knowledge map of public computer course, this method clearly describes the internal relationship between the knowledge points of the course, which is of great significance for guiding the curriculum setting and schedule arrangement and guiding students to learn.

Key words: representation learning, knowledge representation, course management