1.1. What is a social network?
5
degree is likely to be connected to nodes of much lower degree. Many web
clients connect to the web servers at both Microsoft and Google, but these are
much less likely to connect to each other.
• While, from a human perspective, we conceive of ourselves as being members
of multiple groups, such groupings are not, by and large, visible in the aggregate social network if all relationships are treated as the same. If the social
network region around an individual is considered only as a network of relationships, there is no obvious signal that this subregion consists of relatives,
this other subregion of work colleagues, and this third subregion of members
of a club, team, or hobby group. In other words, clusters in a social network are
a perceptual property rather than a structural property, so that finding clusters
or communities in a social network requires some care. This is partly because
there is often substantial overlap among such subregions (a family member
shares an interest in a sport or hobby), and partly because of the presence of
“long” edges.
• Properties that can be considered to flow along the edges of a social network
(that is, properties that are influence-like) travel for surprisingly long distances.
For example, there are some experiments that show that an individual is influenced by the happiness not only of immediate social-network neighbors, but
also by neighbors of neighbors, and even neighbors of neighbors of neighbors
[18, 19]. In general, someone three steps away may not even be known to
the individual, so that these influence-like properties flow “over the horizon”.
Other properties that behave this way include sadness, tendency to smoke, and
being overweight. Unsurprisingly, these effects are of great interest to those
whose business is influence, for example, advertisers and many large-scale
experiments have been enabled by access to internet-scale data [65].
These results show that the social networks in which we find ourselves are not much
like we might expect them to be intuitively. We form relationships with a few other
people based on local information and independent decision making, we build a
mental picture of what the global social network that results is like, but this global
network actually has many properties that are not obvious from the perspective of
any single participant. Emergent structure is the payoff from social network analysis.
Once it is understood, the resulting insights can be used to draw conclusions about
individuals and society that have wide implications. For example, a node may be an
outlier in a social network, that is connected to the network only at the periphery.
This is not the same property as having low degree (although it may be related).
Similarly, a node may be a key node, in some sense, located centrally in the network.
Again, this is not necessarily the same property as having high degree. Emergent
structure of the network as a whole acts as a background against which properties
such as these can become visible.
5
degree is likely to be connected to nodes of much lower degree. Many web
clients connect to the web servers at both Microsoft and Google, but these are
much less likely to connect to each other.
• While, from a human perspective, we conceive of ourselves as being members
of multiple groups, such groupings are not, by and large, visible in the aggregate social network if all relationships are treated as the same. If the social
network region around an individual is considered only as a network of relationships, there is no obvious signal that this subregion consists of relatives,
this other subregion of work colleagues, and this third subregion of members
of a club, team, or hobby group. In other words, clusters in a social network are
a perceptual property rather than a structural property, so that finding clusters
or communities in a social network requires some care. This is partly because
there is often substantial overlap among such subregions (a family member
shares an interest in a sport or hobby), and partly because of the presence of
“long” edges.
• Properties that can be considered to flow along the edges of a social network
(that is, properties that are influence-like) travel for surprisingly long distances.
For example, there are some experiments that show that an individual is influenced by the happiness not only of immediate social-network neighbors, but
also by neighbors of neighbors, and even neighbors of neighbors of neighbors
[18, 19]. In general, someone three steps away may not even be known to
the individual, so that these influence-like properties flow “over the horizon”.
Other properties that behave this way include sadness, tendency to smoke, and
being overweight. Unsurprisingly, these effects are of great interest to those
whose business is influence, for example, advertisers and many large-scale
experiments have been enabled by access to internet-scale data [65].
These results show that the social networks in which we find ourselves are not much
like we might expect them to be intuitively. We form relationships with a few other
people based on local information and independent decision making, we build a
mental picture of what the global social network that results is like, but this global
network actually has many properties that are not obvious from the perspective of
any single participant. Emergent structure is the payoff from social network analysis.
Once it is understood, the resulting insights can be used to draw conclusions about
individuals and society that have wide implications. For example, a node may be an
outlier in a social network, that is connected to the network only at the periphery.
This is not the same property as having low degree (although it may be related).
Similarly, a node may be a key node, in some sense, located centrally in the network.
Again, this is not necessarily the same property as having high degree. Emergent
structure of the network as a whole acts as a background against which properties
such as these can become visible.