Geography Reference
In-Depth Information
5.5 External POI Database
We applied a spatial query on external dataset of geotagged Wiki pages using
coordinates of the area A and B. One page was retrieved for region A: Pays-
d'Enhaut (a district in the canton of Vaud in Switzerland). Three municipalities are
located within the district: Château-d' Œx, Rossinière, and Rougemont) and three
pages for region B: Gürbetal (Gürbe Valley), Historical Carcemetery Gürbetal
(the page exists only in German) and Rümligen Castle (the page exists only in
German). Surprisingly, the Rümligen Castle was not among the attractions of the
region B. However, it turned out that this castle and its adjacent territory is a
private area. Pays-d'Enhaut was not tagged presumably because it represents high
level of abstraction of the area or the majority of the visitors do not know the name
of the district.
We then tried to find relevant information in the web by supplying the extracted
categories to a public web search service. In this way, we managed to obtain the
following information:
1. (Chateau-d'Oex): International Hot-air Balloon Festival takes place every
January.
2. (Kaufdorf): a car cemetery with many old-time cars. It is probable that the car
cemetery existed before summer 2008 but did not attract much attention. On
5th of July, 2008 an article about it was published by the administrator of the
Switzerland group of Flickr users on his web site. This article, evidently,
attracted many Flickr users to visit this place in summer and autumn 2008.
3. (Interlaken): Red Bull Air Races, July 14, 2007.
4. (Axalp): Air Show, October 10-11, 2007.
5. (Thun): Touristic city with several attractions (that do not change over time)
like: the twelfth century Castle, sixteenth century town hall, view of the Alps.
6 Summary
In the use case, we demonstrated the feasibility of the proposed methodology by
applying several approaches of spatio-temporal analysis using geotagged photos
combined with time series analysis, semantic interpretation based on text clus-
tering, and extraction of additional knowledge from external sources. We suc-
ceeded in finding and explaining the events occurring in five selected regions of
Switzerland without any prior background knowledge. The overall runtime per-
formance suggests that the most time consuming operations like spatio-temporal
and text clustering can be applied countrywide without much time overhead.
