Geography Reference
In-Depth Information
2.2.2 Building Gazetteers
Popescu et al. (
2008
) used different Internet sources like Wikipedia, Panoramio
and web search engines to automatically collect, identify and categorize geo-
graphical names.
2.2.3 Working with Photo Collections
An algorithm for summarization of photo collections using textual attributes of a
photo was presented in Jaffe et al. (
2006
). The algorithm, based on Hungarian
method (Kuhn
1955
), first, performed hierarchical clustering of the region using
cluster scoring as a heuristic for cluster creation. The score was composed from
such components as tag-distinguishability, photographer-distinguishability, clus-
ter density, the sum of image qualities. A visualization environment was proposed
to visualize the representative tags for every cluster reflecting the tag's importance.
The later work (Ahern et al.
2009
) used k-means instead of the Hungarian
clustering.
Kennedy et al. (
2007
) applied content and context based analysis for ranking
clusters and finding representative images in a cluster. The cluster ranking was
performed to assess how well the photos in a cluster are represented by a tag. They
included such aspects as number of users, visual coherence, cluster connectivity,
variability in dates. Following, an image analysis was used to select the best
representative image from the high ranked clusters. Image organization and an
engine for discovering landmark photos was proposed in Zheng et al. (
2009b
).
A worldwide landmark list was generated using geotagged images and articles
from travel guides.
Becker et al. (
2009
) proposed an ensemble clustering approach (combining
different features like titles, tags, keywords, description and content creation time)
for event identification (concerts, music festivals, etc.) using photo collections.
Different combinations of features were evaluated where the combination of all
text features and tags alone achieved the highest performance.
3 Our Previous Work
This work is a continuation of a previous work on analysis of event-based
movement data (Andrienko et al.
2009
), visualization of attractive areas using
geotagged photos (Kisilevich et al.
2010a
), and on semantic enrichment of visited
places and pattern mining (Kisilevich et al.
2010b
).
In Andrienko et al. (
2009
), five space and agent-centered analysis tasks for
event and trajectory-based data were defined: spatio-temporal aggregation of
events, spatial clustering of events, spatio-temporal clustering of events, flow
analysis and interactions in space and time.
