With the rapid development of technologies such as big data and artificial intelligence, the next generation of artificial intelligence technologies centered on knowledge graphs has gradually matured, making it possible for teaching knowledge graphs to assist teachers in achieving smart classroom teaching. Based on an introduction to knowledge graph-related technologies and deep learning theories, this paper takes course knowledge as the research object, constructs a course knowledge ontology model and knowledge storage process. Using the constructed course knowledge graph as the foundation, the paper combines stacked LSTM and GCN models to build a course knowledge recommendation model. Stacked LSTM is used for entity-relation extraction, and GCN is used for knowledge mapping, thereby enhancing the effectiveness of course knowledge recommendations. The study found that when the TopN recommendation value was set to 10, the average accuracy of the proposed method was 0.308, which was 1.18 times higher than the UserCF method. Additionally, the knowledge graph, stacked LSTM, and GCN in the model all had a significant impact on improving the performance of course knowledge recommendation. By leveraging deep learning technology, knowledge entities in the data can be better distinguished, thereby facilitating the establishment of a knowledge ontology model and providing a solid foundation for knowledge mapping and recommendation.