GIPSY ( Georeferenced Information Processing SYstem )

In current information retrieval systems, geographically-based searches such as "Find me all documents whose content pertains to location X" are not supported directly by indexing, query, and display functions. Instead, these searches at best exist only as side-effects of keyword indexing. Because of problems relating to ambiguity, synonymy, and other linguistic phenomena, the use of keyword indexing to indicate geographic location can lead to inaccurate and/or misleading references. The initial portion of this paper presents the motivation for georeferencing text and previous work in this area. The next portion discusses a needs assessment presenting two studies documenting the frequency and composition of geographic queries in information retrieval systems. The first study is an examination of commands executed in a university online bibliographic catalog. The second study shows searches done by natural resources managers. The final portion of this paper details a new model for supporting geographicallybased access to text. In this model, words and phrases containing geographic place names and geographic characteristics are extracted from a document. These terms are then used as input to database functions which use spatial reasoning to statistically approximate the location being referred to in the document. Advantages and disadvantages of this model are discussed.