Spatial Information Retrieval with Place Names

Abstract

This work outlines an approach to handle the spatial dimension of information retrieval, i.e. queries of the type concept @ location. It applies a low-level notion of spatial relevance that is closely linked to the location, or geo-reference, of an information object. Methods for a semi-qualitative representation of spatial regions, and for the evaluation of the relative spatial relevance of information objects that are indirectly geo-referenced by place names, are presented. The approach supports a representation of crisp and vague geographic regions that is adequate for most information retrieval tasks. A graph-based abstraction of polygonal tessellations used as referenece system reduces data volumes and supports efficient reasoning algorithms. By combining a semi-qualitative approximation of Euclidean distance with a relative measure for the hierarchical part-of relations of regions into one simple metric, the partonomic semantics of hierarchical place name structures can be leveraged for information retrieval processes.The method was evaluated against a large-scale operational meta-information system (Umweltdatenkatalog - UDK). Implementations include the BUSTER information broker middleware and an OGC compatible catalog-service within a spatial data infrastructure. Other potential application areas include the creation of machine-readable indices of digital maps, the development of spatial reasoning modules for highly distributed ad-hoc (P2P) networks and Semantic Web applications, as well as the integration into new light-weight and location-based applications.