Database Reference
In-Depth Information
imity of one another. However, this cannot necessarily be used in order
to determine whether the corresponding people are interacting with an-
other. A knowledge of such interactions can be determined with the
use of
speech segmentation
techniques in which it is determined which
participants are interacting with one another. The speech portions are
segmented out of the ambient noise, and then segmented into conver-
sations. The knowledge of such face-to-face interactions can be used to
build dynamic and virtual links among the different participants.
We note that a dynamically linked social network can be modeled in
two different ways:
The network can be modeled as a group of dynamic interacting
agents. The stochastic properties of these agents can be captured
with the use of hidden markov models in order to characterize var-
ious kinds of behaviors. This is the approach used for community
modeling as discussed in [15, 36].
The interactions of the participants can be modeled as links which
are continuously created or destroyed depending upon the nature
of the underlying interactions. as a graph stream, in which the
nodes represent the participants, and the edges represent the in-
teractions among these different participants. Recently, a number
of analytical techniques have been designed in order to determine
useful knowledge-based patterns in graph streams [8]. These in-
clude methods for dynamically determining shortest-paths, con-
nectivity, communities or other topological characteristics of the
underlying network.
The inherently dynamic nature of such interactions in an evolving and
dynamic social network leads to a number of interesting challenges from
the perspective of social network analysis. Some examples of such chal-
lenges are discussed below.
(1) Determination of dynamic communities in graph streams:
Communities are defined as dense regions of the social network in which
the participants frequently interact with one another over time. Such
communities in a dynamically evolving social network can be determined
by using agent-based stochastic analysis or link-based graph stream anal-
ysis. Methods for modeling such a social network as a group of dynam-
ically evolving agents are discussed in [15, 36]. In these techniques, a
hidden markov model is used in conjunction with an influence matrix in
order to model the evolving social network.
A second approach is to model the underlying face-to-face interactions
as dynamic links. This creates an inherently dynamic network scenario
in which the structure of the communities may continuously evolve over
