Abstract: Knowledge graph data is a typical network data with inter-entity relationships extracted by natural language processing models from a large amount of text documents, and the major visualization method for network data is using node-link diagrams with force-directed layout. The traditional node-link diagrams with force-directed layout do not take into account the information of entity labels and relationship labels in the knowledge graph data, which leads to the problem of random distribution of entity nodes and relationship links in the layout results. Firstly, the knowledge graph data is preprocessed to obtain the knowledge graph subgraph data with appropriate size. Secondly, 3 new forces are added to the force-directed layout so that the improved layout can better display the relationships between entities, entities and relationship label types in the knowledge graph. Finally, edge bundling techniques are introduced and basic interaction techniques are provided to enhance the visualization and interaction of the method. Compared with other force-directed layout methods, the layout result is more regular in details and more readable on medical knowledge graph data with 20 thousand entities and the cyber blackmail knowledge graph data with 2 million entities.