maps from Flickr geotagged images

above image from here
see this project
text from this Yahoo! Reserach Berkley post
In 1976, social psychologist Stanley Milgram asked his subjects to list places of interest in Paris. Milgram then aggregated the results, effectively creating an “attraction map” of Paris with landmark names appearing in a larger font according to the number of subjects who mentioned each.

Can the same type of information and visualization be automatically derived from Flickr geotagged images and their associated tags? At this year’s ACM MM MIR workshop, we showed the answer to be “yes”. The idea is simple: By taking a photo, photographers essentially express their interest in a particular place. Individual pictures taken at a specific location act as “votes” in favor of that location’s interest, much like the explicit input of Milgram’s subjects. Further, additional information can be extracted from the tags attached to these photos on Flickr. Tags that frequently appear in images from a specific location but are otherwise rare suggest a topic unique to the location.

By analyzing the patterns of location, photographers, and tags in a photo data set, our system generates tag maps that mirror Milgram’s manually created attraction map. While Milgram was testing his ideas in Paris (good for him!) we were looking at London data (we’ll do Paris soon - even if we have to go there and investigate in person). The figure above shows a tag map of central London, derived from Flickr’s geotagged photos. The attractions that emerge from the London data include Buckingham Palace, London Eye, and Big Ben—all generated automatically with the implicit contributions of Flickr users. The data we used to generate the map is Flickr geotagged images from May 2006 - a lot more are available now, and we promise to run it again on the new data set very soon - stay tuned! The beauty of it is that Tag Maps usually improve as more content is added, alleviating the overload problem often associated with large collections.