Exploiting Users’ Map Annotations

Posted: January 19th, 2008 | No Comments »

As part of the Workshop on Volunteered Geographic Information late last year, John Krumm presented a position paper on Exploiting Users’ Map Annotations. In the project, he exploits Live Search Maps’ “Collections” to assess the prominence of existing landmarks and to find new landmarks that should be added to the map. More interestingly, he first mentions the use of this type of data to “determine which landmarks to show at different zoom levels, as a way to describe more obscure locations in terms of prominent ones (e.g. 0.5 kilometers east of Pike Place), and as a way to pick landmarks to give driving directions“. This very similar to part of my work on Tracing the Visitor’s Eye that I presented last year at Ubicomp. Second he talk about an “algorithm for finding new landmarks from collections. In brief, this proceeds by first making geographic clusters of pushpins, extracting all possible one-, two-, and three-word phrases from the associated text, and processing these phrases to find which ones are mentioned frequently in the cluster (e.g. “Space Needle”), but not very often outside the cluster.” I have been developing a similar approach for my project on Florence to reveal the major POI of the city and define their “area of attraction”.

John Krumm is the guest editor of an upcoming special issue of the IEEE Pervasive Computing Magazine on Pervasive User-Generated Content. The CFP introduces the theme as follow:

Pervasively situated people and things can generate content otherwise unobtainable. Small contributions aggregated from a large number of individuals add up to richly minable supply of information and also make the data more credible and reliable. Sometimes the data is generated explicitly, like event reviews, store ratings, and incident reports. Other times, the data comes from analysis of normal behavior, like aggregating traffic speeds from regular drivers. Often people provide the data, like networked games designed to extract everyday knowledge. In other instances, objects like paper currency or ocean-borne, spilled toys are tracked as they move around naturally. The resulting content sometimes comes from sophisticated processing, like inferring driving preferences from observed trips. In other instances, processing is minimal, like text messages displayed publicly in Times Square. In all cases, a critical link is a network that connects the pervasively distributed data sources to a central repository.