Landmark Extraction from Web-Harvested Place Descriptions

Abstract

Large corpora of place descriptions provide abundant human spatial knowledge, different from the geometry-based information stored in current GIS. These place descriptions, used in everyday communication, frequently refer to landmarks. This paper suggests a model for extracting landmarks from web-harvested place descriptions, considering the landmark’s cognitive significance. The model allows landmarks to be extracted according to different contexts via web harvesting and text classification methods. In this work, an implementation of our approach is used to extract context-based landmarks for a target area—Melbourne in Australia.