Georeferencing places from collective human descriptions using place graphs

Place descriptions in everyday communication or in online text provide a rich source of spatial knowledge about places. Such descriptions typically consist of references to places and spatial relationships between them. An important step to utilize such knowledge in information systems is georeferencing the referred places. Beside place name disambiguation, another challenge is that a significant proportion of place references in such descriptions are not official place names indexed by gazetteers, thus cannot be resolved easily. This paper presents a novel approach for georeferencing places from collective descriptions using place graphs, regardless of whether they are referred to by gazetteered names or not. The approach leverages spatial relation models for approximate locating and matching. Different models are proposed and evaluated using several metrics.