Keyword Search in Geospatial Database

摘要

Geospatial databases usually contain features of different geometries, associated with thematic attributes. The features have implicit spatial relationships which are not explicitly captured (unless the relationships have their own attributes, which is rare), and hence geospatial databases contain a few or no foreign keys. As geospatial data becomes more widely available and used by people, their keyword based querying will become an important interface. Keyword queries in geospatial data may be related to features and their attributes, or with multiple features, their spatial relationships, and their attributes. The keywords may refer to feature types as well as feature instances (e.g., "India capital" - here, capital is really a schema level data, i.e., metadata). We must note that keyword queries are not natural language queries. They just contain keywords that the user thinks characterize their expected result. Consequently, analyzing and disambiguating the keyword queries is important before answering them. Spatial relationships to be taken into account may have to be based and ranked on nearest-neighborhood. Finally, the query results need to be shown both textually as well as visually on a map. These requirements make keyword searching in geospatial data quite different from normal data. This paper describes our approach to supporting keyword queries. We describe query interpretation, query translation into one or more SQL queries, and query execution and result ranking. We describe how geo-spatial ontologies can be used in understanding implicit relationships between query objects, and how R-tree based indexes can be effectively used for result ranking based on spatial relationships.