Database Reference
In-Depth Information
detecting places and events from unstructured text, and finding automatic ways
to link data from different sources. These topics are strictly linked to the devel-
opment of the so-called Semantic Web and big companies such as Google and
Yahoo! are constantly researching and developing in these fields. Investigation
of place and event semantics of geo-referenced tags, in addition to the represen-
tativeness, is a prerequisite to using geo-social data. A place tag is defined as a
one that exhibits significant spatial patterns, while an event tag refers to one that
exhibits significant temporal patterns. Both definitions are vague and subject to
some geographic region. For example, “carnival” may not be able to indicate any
event, but will be very specific if only carnivals in NewYork City are considered.
Analyzing the spatial and temporal distribution of tags and identifying the dis-
tributions of events and places with relative geographic scales can be useful to
many applications, such as image search, collection browsing, tag visualization,
and, of course, mobility analysis. Another open issue is the multilingualism of
(geo-referenced) web media. Geo-referenced media is, in fact, multilingual in
nature. However, most systems take English as the sole processing language.
This effectively excludes the media resources in other languages. The conse-
quence is that the knowledge and patterns mined from geo-referenced media are
biased toward English-speaking countries and regions, though people are more
comfortable using their local language (also dialects and slang) to communicate
with friends, especially in colloquial sentences such as the ones used in chats,
SMS, or status updates, or in stressful, demanding situations such as disasters or
danger. The geographic locations of photos on the Internet have opened up a new
host of research and application possibilities. As described in the photo example
in Section
16.4.4
, a spatial gap can exist between the GPS camera position and
the position of the subject in the photo. Knowing the geographic orientation of
photos, that is, in which direction the cameras are pointing, will be useful to
fill the gap. Though most cameras are not equipped with sensors to measure
the orientation and inclination of the device, smart photos, with the iPhone
and HTC Magic as prime examples, have started to embrace digital compass
technologies. In addition to hardware sensors, software solutions to estimate
photo orientation also exist, for example estimating the relative translation and
orientation between photos, by leveraging the visual redundancy among photos.
Till now, geographic orientation of photos was rarely available. Nevertheless,
with the development of compass-equipped cameras and smartphones, such kind
of metadata is expected to emerge in the near future. With the availability of
photo orientation metadata, many compelling applications can be accomplished.
For example, with the photo alignment information, visual summarization and
browsing of photo collections can be adaptive to the user direction and perspec-
tive on the map. Moreover, 3D reconstruction of geolocation can be much more
efficient.
