An Abstraction-based Machine Learning Approach to Cartographic Generalisation

This article proposes a machine learning approach to overcome the knowledge acquisition bottleneck limiting the automation of cartographic generalisation. It first explains why this automation must be guided by a differentiation of two main types of knowledge involved in this process. More precisely, it shows that cartographic generalisation is best viewed as a combination of two processes: representing (formulating, renaming knowledge) and abstracting (simplifying a given representation). The whole process of creating maps fits into an abstraction framework developed in Artificial Intelligence to account for the difference between knowledge abstraction and knowledge representation. The utility of this framework lies in its efficiency to support the automation of knowledge acquisition for cartographic generalisation as a combined learning of both abstraction and representation knowledge. The results of Machine Learning experiments show the advantages of this approach.