The knowledge question-answering system leverages knowledge graphs for retrieval and inference, enabling it to handle complex queries. However, knowledge graphs in specialized technical fields are often incomplete, with limited coverage, and fail to provide sufficient support for question-answering systems. To address these issues, this paper delves into relevant knowledge graph technologies and proposes a node classification and link prediction method based on a hybrid teacher model. This method consists of multiple teacher models, integrating the strengths of different teacher models’ knowledge. Based on this foundation, tasks for node classification and link prediction have been developed.In addition to analyzing each module’s overall results, experiments on citation network datasets and similar datasets verify that the model performs better in terms of network classification and link prediction tasks across many datasets.