Geographic knowledge graphs play a significant role in the geospatial semantics paradigm for fulfilling the interoperability, the accessibility, and the conceptualization demands in geographic information science. However, due to the immense quantity of information accompanying and the enormous diversity of geographic knowledge graphs, there are many challenges that hinder the applicability and mass adoption of such useful structured knowledge. In order to tackle these challenges, this dissertation focuses on devising ways in which geographic knowledge graphs can be digested and summarized. Such a summarization task, on the one hand lifts the burden of information overload for end users, on the other hand facilitates the reduction of data storage, speeds up queries, and helps eliminate noise. The main contribution of this dissertation is that it introduces the general concept of geospatial inductive bias and explains different ways this idea can be used in the geographic knowledge graph summarization task. By decomposing the task into separate but related components, this dissertation is based upon three peer-reviewed articles (Chapter 3, Chapter 4, and Chapter 5) which focus on the hierarchical place type structure, multimedia leaf nodes, and general relation and entity components respectively. Chapter 6 presents a spatial knowledge map interface to illustrate the effectiveness of summarizing geographic knowledge graphs. Throughout the dissertation, top-down knowledge engineering and bottom-up knowledge learning methods are integrated. We hope this dissertation would promote the awareness of this fascinating area and motivate researchers to investigate related questions.