Ramaswamy Hariharan, John Krumm, and Eric Horvitz
Location-based services like reminders, electronic graffiti, and tourist guides normally require a custom, location-sensitive database that must be custom-tailored for the application at hand. This deployment cost reduces the initial appeal of such services. However, there is much location-tagged data already available on the Web which can be easily used to create compelling location-aware applications with almost no deployment cost. Such tagged data can be used directly in applications as well as to provide evidence in models of ac-tivity. We describe three applications that take advantage of existing Web data combined with location measurements from a GPS receiver. The first applica-tion, “Pinpoint Search”, finds web pages of nearby places based on GPS coor-dinates, queries from a Web mapping service, and general Web searches. The second application, “XRay”, uses the mapping service to find businesses in a building by pointing a GPS-equipped electronic compass at the building. The third application is called “Travelogue”, and it builds a map and clickable points of interest to help automatically annotate a trip based on GPS coordi-nates. Finally, we discuss the use of Web-based data as rich sources of evidence for probabilistic models of a user’s activity, including a means for interpreting the explanation for the loss of Web signals as users enter structures.