Geography Reference
In-Depth Information
The procedure for the visualization of attractive areas was proposed in
Kisilevich et al. (
2010a
). The process consists of applying a density based clus-
tering algorithm to the photo data and calculating the importance score of a photo
using kernel density estimation. The importance score was then used for two
purposes: (1) as a value for colour generation and (2) an estimator for importance
of the photo. The photo with the largest importance score was selected as a
representative photo in a cluster.
A four-step process was proposed in Kisilevich et al. (
2010b
) to extract
movement sequence patterns using a semantic enrichment process. In the first step,
every photo was semantically annotated by a nearest point of interest (POI) using
an external database of POIs. The photos that were assigned to the same POI
created a semantic cluster with the POI being a representative of the cluster. For
example, if the POI is a train station, the question can be asked: Are there people
who take photos near a train station or how many people take photos near a train
station?
In the second step, photos that were not semantically annotated due to the
absence of POI in the neighborhood, were clustered into regions. The obtained
regions were considered as new unknown POIs. In the third step, a movement
sequence was generated for every user, using the POI identifiers assigned to her
photos. In the fourth step, a sequence mining algorithm was applied to the
sequences in order to find frequent patterns. As a consequence, the pattern of type
A ? B could be interpreted like this: people who visit the area A also visit the
area B and pattern of type
A ? * ? B could be interpreted like: people who visit the area A may con-
tinue to any other place and from any other place come to B.
The current paper extends the previous work in several aspects:
(1) To reflect the importance of time in cluster creation and analysis, we provide a
taxonomy of possible types of spatio-temporal clusters.
(2) The semantics enrichment process is discussed with respect to time.
(3) We discuss the possible external data sources that can facilitate extraction of
semantics.
(4) The methods supporting semantics extraction are outlined.
4 Importance of Time in Understanding Space
In
Sect. 1
we argued that time is important for understanding spatial patterns.
In this section, we provide a taxonomy of spatio-temporal clusters and present
possible data sources of semantics knowledge. Additionally, we discuss methods
that facilitate semantic extraction and understanding of spatio-temporal clusters.
