Abstract: Knowledge graph technology has developed rapidly in recent years. However, the research on domain knowledge graph construction is preliminary. There are mainly two methods to construct a knowledge graph, one is the topdown, and the other is the bottomup. The former method is more common. However, the topdown has a disadvantage, that ontology learning is too artificial, so that it does not support the construction of a domain knowledge graph which is strong professional and massive. In order to solve this problem, this paper proposes a novel domain knowledge graph construction method, which combining the topdown and the bottomup. We also propose an improved skeletal methodology to ensure the accuracy and coverage in ontology construction. We use rulesbased and semisupervision technology to improve the speed of knowledge extraction. After that, we illustrate the procedure of construction in detail. Moreover, instances prove that the method in this paper is feasible.