Making Sense of the World, One Photo at a Time
             Mor Naaman, Yahoo! Research Berkeley

The availability of map interfaces and location-aware devices makes a
growing amount of unstructured, geo-referenced information available
on the Web. This type of information, in aggregate form, can help
understand data trends and features. In particular, over ten million
geo-referenced photos are now available on the photo-sharing website
Flickr. These photos are often associated with user-entered
unstructured text labels (i.e., tags) -- the first major collection of
its kind.

In this talk, I will discuss two approaches for extraction of
information (knowledge, if you will) from this metadata-rich yet
unstructured set of photos. The first approach is location-driven,
using the dataset to extract representative tags for each map region
and zoom level. An alternative approach takes a tag-centric view, and
attempts to extract place/event semantics for each tag by analyzing
the usage patterns of the tag in location and time.

Finally, I will describe and demonstrate two prototype applications
that make use of the extracted knowledge, ZoneTag and TagMaps, both
available from the Y!RB web site at http://whyrb.com.