Database Reference
In-Depth Information
geographic factors and social mechanisms. Cranshaw et al. (2010) studied the
entropy related to LBSNs locations in order to understand how it affect the
underlying social network. They found that co-locations at high entropy loca-
tions are much more likely to be random occurrences than co-locations at low
entropy locations. So, if two users are only observed together at locations of
high entropy (for example, a shopping mall or a university), they are less likely
to actually have a link in the underlying social network than if they are observed
in a place of low entropy. Moreover, users who visit locations of higher entropy
tend to be more social, having more ties in the social network than users who
visit less diverse locations.
Trajectory Overlap
Given that two persons have been on multiple occasions in the same geographic
place at the same time, how likely are they to know each other? This is another
interesting and open problem about the interplay between sociality and mobility,
regarding to which extent social ties between people can be inferred from co-
occurrence in time and space.
Crandall et al. (2010) studied this problem by analyzing a huge data set from
the popular photo sharing site Flickr, reaching interesting and striking conclu-
sions. They inferred a spatio-temporal co-occurrence between two Flickr users
if they both took photos at approximately the same place and at approximately
the same time. Rather surprisingly, they found that even a very small number of
co-occurrences can lead to orders-of-magnitude greater probabilities of a social
tie. Indeed, two users have nearly 5,000 times the baseline probability of having
a social tie on Flickr when they have just five co-occurrences in a day in an
80-km range of distance. With the aim of a deeper understanding of the under-
lying phenomenon, they developed a mathematical model in which the proba-
bilities of friendship as a function of co-occurrence qualitatively approximate
the distributions they observed in the Flickr data.
Wang et al. (2011) presented a data-mining approach to the question of to
what extent individual mobility patterns shape and impact the social network.
Following the trajectories and communication patterns of approximately 6 mil-
lion mobile phone users over 3 months, they defined three groups of similarity
measures: mobile-homophily (similarity in trajectories), network proximity (dis-
tance in the call graph), and tie strength (number of calls between two users).
Exploring the correlation between these measures, researchers discovered that
they strongly correlate with each other. The more similar two users' mobil-
ity patterns are, the higher the chance that they have close proximity in the
social network, as well as the higher the intensity of their interactions. Starting
from these results, they designed a link prediction experiment, constructing the
entire repertoire of both supervised and unsupervised classifiers, based either
on network and/or mobility quantities. Results showed that mobility on its own
