Geography Reference
In-Depth Information
Thus, if we know that a photo is visually similar to a nearby photo, then these two
photos can be grouped together. If needed, the representative photo can then be
found. Content-based analysis can also reveal the heterogeneity of the cluster. If
people take photos of a single point of interest, then a large amount of photos will
be similar to each other.
4.3.4 Analysis of Time Series of Counts of Events
Time series analysis helps to understand the type of the cluster in terms of its
temporal characteristics. For example, we can count the number of people who
take photos in the area every day within a 1 year period. The shape of the time
series indicates the type of a cluster. If the number of events does not differ
significantly, the cluster can be classified as stationary. If there are bursts in
activity at several intervals, the cluster can be classified as reappearing. If there is a
single high peak, the cluster can be classified as occasional.
Additionally, the change of contexts can be detected by using time series
analysis. Following the context detection using text and content based analysis, we
can classify clusters to any of the spatio-temporal types that we already defined in
terms of number and variety of events occurring in a cluster during some time
period. Let us consider an example: There is a cluster in which sport events such as
alpine skiing are held in winter. In the summer time, most of the photographers
take pictures of nature. In this case, we can classify the cluster as reappearing if the
context (winter sport, nature photography) appears in the cluster several times. If
the context appears only once, it is probable that the cluster is occasional.
4.3.5 POI Database and Entity Relations
POI database is a valuable source of semantics. The methods used in automatic
gazetteer construction may be employed in retrieving the topic of the photo (the
title on the photo relates to a landmark or nature) or build an ontology of types
from thematic photo groups. In this case, the photos that are in the group birds can
be classified in general as being of type nature.
4.3.6 User Profiling
As was shown above, cluster semantics may help to determine the semantics of
user behavior and vice versa. The user profiling can be performed statically, by
using information from all the photos that belong to the user or dynamically, for
every cluster or for different time periods.
