Database Reference
In-Depth Information
can distinguish between a check-in to a different floor of the same building.
To give an example, Yahoo! research labs published a study on the attempt to
extract aggregate knowledge on certain locations from large scale geo-referenced
photos at Flickr. The knowledge here refers to the word or concept that can
best describe and represent a geographical region. The challenge is to extract
structured knowledge from the unstructured set of tags. The premise of the
proposed solution is based on the human attention and behavior embedded in
the photos and tags. Namely, if tags concentrate in a geographical area but do
not occur often outside that area, then these tags are more representative to
the area than those spread over large spatial region. This example shows also
that there is a need to model human behavior and this aspect constitutes an
interesting research topic by itself. Of course, models and hypotheses are geo-
graphically dependent as Western people often act differently from Eastern peo-
ple in a social context. However, online social networks' check-ins are usually
more sporadic than phone calls, providing less temporal resolution than mobile
data.
Some references for further information are provided in Section
16.7
.
Theoretical Application Scenarios
In this section we describe some possible scenarios where the analysis of virtual
movements in geo-social networks can be useful, but has not yet been investi-
gated by researchers. For instance, in the emerging field of human dynamics,
a central point is the understanding of the interplay between human mobility
and social networks. How do the mobility patterns and parameters depend on
social network characteristics? The study of such interaction requires massive
society-wide data sets that simultaneously capture dynamical information on
individual movements and social relationships. Traditionally, this problem is
addressed by using mobile phone networks, because they provide at the same
time temporal information and social contacts. However, there are at least two
problems with this kind of mobile phone data. First, friendships are not explicit
but are inferred by creating a who-called-whom graph, with the possibility of
inaccurate information about tie strengths. For example, a person does not often
call people who live with him or her. The low number of calls between them is
interpreted as a weak tie, leading to a bad representation of reality. This aspect
is overcome by social networks where strong ties generally generate more direct
messages/interactions. Secondly, we know users' positions only when they per-
form a call, and merely know the position of the tower managing the area the
user is within, and not the actual geographical location of the user. In the geo-
social network application the user's location can be retrieved when he or she
publishes content and can also be derived from the user's, friends' contents if
he or she is moving with them. In the last case some level of uncertainty is
introduced (see Section
16.4.4
).
