159
reveal many subtle properties about the social network being modelled. At their
simplest, the length of such an edge in the embedding measures the discrepancies
between the individual’s role in the network described by each layer. This can be
used to detect that an individual acts quite differently in a social setting than in a
work setting, or that an individual has suddenly changed his or her role in a group
from one time period to the next.
In general, the length of an embedded edge “should” match its weight — edges
with large weights should be shorter than edges with small weights. In a perfectly
consistent social network, the ratio of embedded length to weight should be constant. Therefore, edges for which this is not true are of special interest — they are
being pushed or pulled from their natural location by the nodes in their neighborhood. Computing measures based on discrepancies between expected and actual
embedded length provide a way to focus attention on the interesting pieces of large
social networks, those for which visualizations are too complex and cluttered for
easy direct analysis.
We have also shown that the layered approach can be composed to model edges
with more than one kind of semantics at once. This means that the layered approach
can be used to build up models of increasing complexity by adding new edge types
one at a time. This is not an automatic process — the difficulty is to choose the
weights for the added edges in a principled way as the number of possible connections in the nexus of multiple versions of each individual node increases.
Discrepancies between expected and actual embedded length are particularly
useful for the extra edges added in the nexus when a composed layered model is
used. These edges represent, as before, discrepancies between the roles of the individual associated with that node with respect to all of the edge semantics in play
in the layered composition. Especially as the number of versions of each node almost guarantees a cluttered visualization, calculating the discrepancies for all nodes
is a way to focus attention on those parts of the social network that are especially
interesting.
We have shown how these techniques can be applied to real-world data of
varying kinds. Often, the differences between existing techniques and the layered
approach make it clear why the layered approach is an improvement.
reveal many subtle properties about the social network being modelled. At their
simplest, the length of such an edge in the embedding measures the discrepancies
between the individual’s role in the network described by each layer. This can be
used to detect that an individual acts quite differently in a social setting than in a
work setting, or that an individual has suddenly changed his or her role in a group
from one time period to the next.
In general, the length of an embedded edge “should” match its weight — edges
with large weights should be shorter than edges with small weights. In a perfectly
consistent social network, the ratio of embedded length to weight should be constant. Therefore, edges for which this is not true are of special interest — they are
being pushed or pulled from their natural location by the nodes in their neighborhood. Computing measures based on discrepancies between expected and actual
embedded length provide a way to focus attention on the interesting pieces of large
social networks, those for which visualizations are too complex and cluttered for
easy direct analysis.
We have also shown that the layered approach can be composed to model edges
with more than one kind of semantics at once. This means that the layered approach
can be used to build up models of increasing complexity by adding new edge types
one at a time. This is not an automatic process — the difficulty is to choose the
weights for the added edges in a principled way as the number of possible connections in the nexus of multiple versions of each individual node increases.
Discrepancies between expected and actual embedded length are particularly
useful for the extra edges added in the nexus when a composed layered model is
used. These edges represent, as before, discrepancies between the roles of the individual associated with that node with respect to all of the edge semantics in play
in the layered composition. Especially as the number of versions of each node almost guarantees a cluttered visualization, calculating the discrepancies for all nodes
is a way to focus attention on those parts of the social network that are especially
interesting.
We have shown how these techniques can be applied to real-world data of
varying kinds. Often, the differences between existing techniques and the layered
approach make it clear why the layered approach is an improvement.