Database Reference
In-Depth Information
simultaneously derived and used [52] for inferring the significant links.
The work in [44] derives the links between users based on their mobility
patterns from GPS trajectories. In order to achieve this goal, the work in
[44] divides the spatial regions into a grid, and constructs nodes for each
cell. An edge exists between a pair of nodes, if a trajectory exists which
starts at one cell and ends at another. By performing the discretization
at varying levels of granularity, it is possible to analyze different char-
acteristics of the underlying users. The work in [44] specifically shows
how the approach can be used for effective community detection.
An interesting work in [173] examines the common patterns in the
activities of different geo-tracked users, and makes friendship or linkage
recommendations on the basis of significant overlaps in activity patterns.
It has also been observed in [115] that different kinds of sharing in ac-
tivity patterns may have different significance for different users. For
example, it is possible that two individuals that are friends may not
spend a lot of time together, but only a couple of hours on a Saturday
night. On the other hand, a pair of co-workers who are not friends may
share a lot of time together. Thus, it is critical to be able to learn the im-
portance of different kinds of commonality in patterns in the prediction
process [115]. Such trajectory analysis is useful not just for determin-
ing useful relationships, but also interesting places, travel sequences or
activities which are relevant to such relationships [27, 181]. In particu-
lar, an interesting
authority-based
model for relating social behavior and
location behavior has been proposed in [27]. The essential idea is to
construct a graph which models relationships of the trajectories of the
different users to the different locations. The idea is that authoritative
users are also likely to visit authoritative places and vice-versa. This is
used in order to construct a page-rank like model in order to determine
both the authoritative users and authoritative locations simultaneously.
Many sensing platforms such as those discussed in [33], yield sensor
data which is varied, and is of a multi-modal nature. For example, the
data could contain information about interactions, speech or location.
It is useful to be able analyze such data in order to summarize the in-
teractions and make inferences about the underlying interactions. Such
multi-modal data can also be leveraged in order to make predictions
about the nature of the underlying activities and the corresponding so-
cial interactions. This also provides a virtual technique to perform link
inferences in the underlying network.
The collection of activity sensing data is not very useful, unless it can
be leveraged to summarize the nature of the activities among the differ-
ent participants. For example,in the case of the techniques discussed in
[34], the IR transceiver is used to determine which people are in prox-
