2.3 Spatial Centrality Measures of A Node

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Spatial Centrality Measures of a Node

In the last article, we discussed two different centrality measures, namely the
eigenvector centrality and the page rank. These centrality measures are useful for
finding important nodes in order to advertise in social networks or for ranking websites
on the world wide web, but they're less appropriate for other applications.
So in this lecture, we'll discuss other centrality measures as well as their applications.
→Let's consider the problem of deciding in which city to build a new airport.
You would first map out all airports that are close by and determine the region that the
new airport would serve.

Then it will be reasonable to choose the location of the new airport so that it is most
central in terms of average travel time to and from the airport to the cities in the region.
So, in this application, the vertices are the cities in the region. And the undirected
edges carry edge weights, namely the travel time between two cities.

1
So the adjacency matrix is symmetric. And the entries of the matrix are zero if there is
no direct street connecting two cities without passing through a third city. Otherwise,
the value in the adjacency matrix are equal to the travel time. Although there are
usually many possible ways to get from one city to another, it is reasonable to assume
that people take the shortest route. This is also called the Geodesic path. So the goal
of this application is to find the city that has the shortest average Geodesic distance to
all other cities.
Thinking about airport location, you might want to appropriately scale the distances by
the size of the cities. This can be done, but it's easier to avoid scaling things the first
time around.
→So how do we compute these distances from the adjacency matrix?
We've already discussed how to compute the lengths of paths between any two
nodes. For example, the adjacency matrix squared contains the travel times between
any two cities that cross through a third city. More generally, the adjacency matrix AK
contains the travel times between two cities on a path that goes through (K -1) cities.
So the geodesic distance of city I and city J is the minimum nonzero value of the I-J
entry of the adjacency matrix A to some power K range over all K’s.
So this is another very nice example of how we can use the adjacency matrix to gain a
lot of useful information. To end the discussion about where to build a new airport,
let's denote by DIJ the geodesic distance between city I and city J.So for each city, we
can compute its average geodesic distance to all other cities. If we have N cities, then
this is one divided by N-1 times the sum of the geodesic distances to all other cities,

And we will choose to build the airport in the city with the smallest average.
This measure of the importance of a node is called closeness centrality.
And similarly to the eigenvector centrality discussed during the last article, there are
various variances of the centrality measure. To give another example, we all know how
quickly rumors spread and in particular how quickly the message in a rumor gets

2
distorted. If the goal is to reach as many people as possible with an accurate message,
a reasonable strategy might be to start spreading the message from a person with a
high closeness centrality who on average has a shorter geodesic distance to all other
members of the community.
So here we discussed closeness centrality and different applications where one might
be interested in the centrality measure.
However, in other applications, we might not be interested in identifying the most
centrally located node in terms of distances. We might instead want to identify
possible bottlenecks of the network.

→ So how should we define a bottleneck?


So a bottleneck can be defined as a node whose removal
from the network will most disrupt the information flow in
the network. Equivalently a bottleneck is a node through
which the largest quantity of information flows. We can
identify such nodes by something called the betweenness
centrality.
The betweenness centrality makes various assumptions
to approximate the amount of information that flows
through each node. It assumes that information flows at
constant speed along the edges and that information
takes the shortest path.
Then the amount of information that passes through
each node is proportional to the number of geodesic
path that goes through that node. This measure is called
the betweenness centrality.

So we've already seen how to compute the geodesic paths. Now a very
efficient algorithm to compute the shortest paths between all pairs of
nodes is a so-called Floyd-Warshall algorithm. It's an example of
dynamic programming. So the algorithm is based on the following idea.
Let's start with just 3 nodes. So here it's easy to find the shortest paths
between any two nodes. Let’s save these paths.

3
Now we add the fourth node. So now the shortest path
between nodes one and two is either the shortest path that we
already found in the small graph or there is a new shortest path
that also uses node 4.
That path can be written as a sum of two paths. Namely one
from node 1 to node 4 possibly using node 3 and then one from
node 4 to node 2 possibly using node 3.

So you can see how you can find the shortest paths in larger graphs by only knowing
the shortest paths in smaller sub-graphs. In many applications, it is important to be
able to find nodes with high betweenness centrality. In social networks, for example,
these are the people who have a lot of power to change messages that are being
passed around. In transportation networks, such as an airline flight network, these
nodes, or in this case, airports, are the most vulnerable nodes for attacks and need to
be well protected.
Until now, we've only talked about centrality measures of vertices, but we could
equally well talk about centrality measures of edges. For example, in many
applications, such as road networks or power grids, it is important to determine the
edges that could be a bottleneck. The importance of an edge can be defined
equivalently to the betweenness centrality of a node by the number of geodesic paths
that passed through this edge.
So to summarize, in this article, we have discussed two additional measures of
centrality, namely the closeness centrality and the between the centrality. Centrality
measures can be used to identify important nodes or edges in the network.

You might also like