With the continuous improvement of digital construction of government affairs, the need for realizing “knowledge-based, personalized and intelligent” government services based on knowledge graph is gradually being awakened. At present, the application of knowledge graph in the government domain is often oriented to a single scenario, making it difficult to establish connections between different scenarios with respect to government knowledge. The search, management and approval efficiency of government services based on traditional databases is still not high. To expand the scope of government services and improve the efficiency of search, management, examination and approval, this paper proposes a set of top-down mapping methods to construct a multi-layer government knowledge graph. Specifically, this method first construct the conceptual model of government knowledge from the perspective of government service, and then obtain the government knowledge, data preprocessing and knowledge fusion according to the conceptual model; finally, forming a multi-layer government knowledge graph with top-down relationships of concept, business service, social service and information sharing. With the visual display of Neo4j and the deployed services, paper validate the proposed method by taking the search of real estate approval, the examination and approval of occupied forest land and the social service for public complaints as examples, which proves that the method is efficient and feasible. It not only provides knowledge graph support and association for different government scenarios, but also helps to realize the fusion and sharing of multi-source government data, thus providing a reference for the subsequent construction of government knowledge graph.