Custom local search.

Many popular online services provide "local" or "yellow-pages" search, but none of them allow users to customize the search over user-specified data. This paper describes a novel system for providing custom local search over user-provided spatial datasets logically combined with detailed street-level vector data. We present the algorithms and architecture underlying our prototype system, which handles complex queries and elegantly solves the problem of parsing terms from multiple datasets in the presence of ambiguity and misspellings. We show that the system provides custom local search with precision and recall figures that often exceed that of a commercial local search service. Further, we show that our system scales gracefully with the number of individual custom data sets being served simultaneously.