Geography Reference
In-Depth Information
5 Case Study: Using Time-Series Analysis and Text
Clustering for Extracting Semantics of Events and Places
In this section, we demonstrate analysis of temporal patterns and semantic
acquisition using combination of time series, text analysis and external data
sources presented in
Sect. 4
.
5.1 General Scenario
Let us briefly consider a possible scenario by employing the methods presented in
Sect. 4
in the analysis of a geographic region.
1. We apply a clustering algorithm to outline areas of people's visits. Although the
cluster and its size reveal spatial information, it explains neither the dynamics
of the interest nor why the place was interesting to the photographers. There-
fore, additional investigation should be performed.
2. We apply time-series analysis to investigate peaks of activity. The dynamics of
the subject of interest can change over time and the same cluster can encompass
different events that also change over time. The temporal component of the
semantic enrichment will change the way we analyze spatio-temporal processes
and as a result, different patterns of spatio-temporal clusters will appear. The
number of taken photos or number of people can be used as the dependent
variable. At this level, we can already infer the spatio-temporal type of the
cluster according to the selected dependent variables. While peaks of activity
can point to some interesting time periods, we still cannot deduct what was the
reason of such activity.
3. We apply clustering techniques for extracting significant keywords using photo
tags and/or titles that can show the photographers' intended subjects and points
of interest when taking the photos. In fact, text clustering techniques can be
applied on all the photos in a cluster or separately on photos for each time
interval. This approach can reveal changing trends of place interest.
4. We can use external POI databases like Wikipedia to acquire additional
information about the cluster if there are points of interest in the area. This
information can be matched against the topics obtained from the text clustering
step.
5. We can apply image clustering to find representatives that visually highlight the
place or in cases where the text clustering does not provide meaningful
categories.
6. Methods used in gazetteer creation supported by the domain expert can be
employed in building hierarchies of concepts for the given cluster, e.g. a photo
of an animal will be classified as nature. Retrieval of other places with similar
events can be performed by searching for areas with similar semantics.
