Landmark Extraction: A Web Mining Approach

Landmarks play crucial roles in human geographic knowledge. There has been much work focusing on the extraction of landmarks from geographic information systems (GIS) or 3D city models. The extraction of landmarks from digital documents, however, has not been fully explored. The World Wide Web provides a rich source of region related information based on our understanding of geographic space. Web mining enables a new mean of extracting landmarks, differently from conventional vision oriented methods. Our approach is based on how geographic objects are expressed by humans, instead of how they are observed. We extend existing methods of text mining so that spatial context is considered. The results of the experiments showed that adopting spatial context into text mining improves the precision of extracting landmarks from web documents.