Node

Centrality

Pedro Ribeiro
(DCC/FCUP & CRACS/INESC-TEC)

(Heavily based on slides from Jure Leskovec and Lada Adamic @ Stanford University)
 Star Wars IV Network
moviegalaxies.com

Are all nodes “equal”? How to measure their importance?

Pedro Ribeiro – Node Centrality
Star Wars IV Network

Size proportional to degree: is this the only way?

Pedro Ribeiro – Node Centrality
Star Wars IV Network

Size proportional to betweenness

Pedro Ribeiro – Node Centrality
Star Wars IV Network

Size proportional to closeness

Pedro Ribeiro – Node Centrality
Why degree is not enough

Pedro Ribeiro – Node Centrality

Why degree is not enough

Pedro Ribeiro – Node Centrality

 Different notions of centrality
●
Node Centrality measures “importance”

Pedro Ribeiro – Node Centrality

 Node Degree
●
Let’s put some numbers to it
Undirected degree:
e.g. nodes with more friends are more central.

Assumption: the connections that your friend has don't

matter, it is what they can do directly that does (e.g. go
have a beer with you, help you build a deck...)

Pedro Ribeiro – Node Centrality

 Node Degree
●
Normalization:
divide degree by the max. possible, i.e. (N-1)

Pedro Ribeiro – Node Centrality

 Node Degree
example financial trading networks

high in-centralization: low in-centralization:

one node buying from buying is more evenly
many others distributed

Pedro Ribeiro – Node Centrality

 What does degree not capture?
●
In what ways does degree fail to capture
centrality in the following graphs?

Pedro Ribeiro – Node Centrality

Brokerage not captured by degree

Pedro Ribeiro – Node Centrality

Brokerage: Concept

Pedro Ribeiro – Node Centrality

Brokerage: Concept

Pedro Ribeiro – Node Centrality

 Capturing Brokerage
●
Betweenness Centrality:

intuition: how many pairs of individuals would have to

go through you in order to reach one another in the
minimum number of hops?

Pedro Ribeiro – Node Centrality

 Betweenness: Definition

g jk (i)
C B (i)=∑
j<k g jk
Where:
gjk = the number of shortest paths connecting nodes j and k
g jk(i) = the number that node i is on.

Usually normalized by:

' C B (i)
C (i)=
B
(n−1)(n−2)/2
number of pairs of vertices
excluding the vertex itself

Pedro Ribeiro – Node Centrality

 Betweenness: Toy Networks
●
Non-normalized version:

Pedro Ribeiro – Node Centrality

 Betweenness: Toy Networks
●
Non-normalized version:

– A lies between no two other vertices

– B lies between A and 3 other vertices: C, D, and E
– C lies between 4 pairs of vertices: (A,D),(A,E),(B,D),(B,E)
●
note that there are no alternate paths for these pairs
to take, so C gets full credit

Pedro Ribeiro – Node Centrality

 Betweenness: Toy Networks
●
Non-normalized version:

Pedro Ribeiro – Node Centrality

 Betweenness: Toy Networks
●
Non-normalized version:

●
why do C and D each have
betweenness 1?
●
They are both on shortest
paths for pairs (A,E), and (B,E),
and so must share credit:
- 1⁄2+1/2 = 1

Pedro Ribeiro – Node Centrality

 Betweenness: Toy Networks
●
Non-normalized version:

What is the betweenness

of node E?

Pedro Ribeiro – Node Centrality

 Betweenness: Real Example
●
Social Network (facebook)
nodes are sized by degree, and colored by betweenness

Pedro Ribeiro – Node Centrality

 Betweenness: Question
●
Find a node that has high betweenness
but low degree

Pedro Ribeiro – Node Centrality

 Betweenness: Question
●
Find a node that has low betweenness but
high degree

Pedro Ribeiro – Node Centrality

 Closeness Centrality

●
What if it’s not so important to have
many direct friends?

●
Or be “between” others

●
But one still wants to be in the “middle”
of things, not too far from the center

Pedro Ribeiro – Node Centrality

 Closeness Centrality
●
Need not be in brokerage position

Pedro Ribeiro – Node Centrality

 Closeness: Definition
●
Closeness is based on the length of the
average shortest path between a node
and all other nodes in the network

Closeness Centrality:
1
C C (i)= N

∑ d (i , j)
j=1

Normalized Closeness Centrality:

'
C (i)=C C (i)×(n−1)
C
When graphs are big, the
-1 can be discarded and
we multiply by n
Pedro Ribeiro – Node Centrality
Closeness: Toy Networks

Pedro Ribeiro – Node Centrality

Closeness: Toy Networks

Pedro Ribeiro – Node Centrality

 Closeness: Question
●
Find a node which has relatively high degree
but low closeness

Pedro Ribeiro – Node Centrality

 Closeness: Question
●
Find a node which has low degree but
high closeness

Pedro Ribeiro – Node Centrality

 Closeness: unconnected graph
●
What if the graph is not connected?

1
C C (i)= N

∑ d (i , j)
j=1

instead of null, we could also interpret

it as 0 if infinity is the distance
between unconnected nodes

Pedro Ribeiro – Node Centrality

 Harmonic: Definition
●
Replace the average distance with the harmonic
mean of all distances

Harmonic Centrality:
1 1
C H (i)=∑ = ∑
j≠i d (i , j) d(i , j)<∞ , j≠i d (i , j)
– Strongly correlated to closeness centrality
– Naturally also accounts for nodes j that cannot reach i
– Can be applied to graphs that are not connected

Normalized Harmonic Centrality:

'
C (i)=C H (i)/(n−1)
H
Pedro Ribeiro – Node Centrality
 Harmonic: Toy Networks
●
Non-normalized version:

Pedro Ribeiro – Node Centrality

 Closeness vs Harmonic

Closeness Centrality Harmonic Centrality

1 1
C C (i)= N C H (i)=∑
∑ d (i , j) j≠i d (i , j)
j=1

Pedro Ribeiro – Node Centrality

 Eigenvector Centrality
●
How “central” you are depends on how “central”
your neighbors are

Pedro Ribeiro – Node Centrality

 Eigenvector Centrality
Eigenvector Centrality:
n
1
C E (i)= ∑ A ji ×C E ( j)
λ j=1
where λ is a constant and
Aij the adjacency matrix (1 if (i,j) are connected, 0 otherwise)

(with a small rearrangement) this can we rewritten

in vector notation as in the eigenvector equation
Ax= λ x
where x is the eigenvector, and its i-th component is the centrality of node i

In general, there will be many different eigenvalues λ for which a non-zero eigenvector solution exists. However,
the additional requirement that all the entries in the eigenvector be non-negative implies (by the Perron–Frobenius
theorem) that only the greatest eigenvalue results in the desired centrality measure

Pedro Ribeiro – Node Centrality

Bonacich eigenvector centrality
also known as Bonacich Power Centrality

Pedro Ribeiro – Node Centrality

Bonacich eigenvector centrality
also known as Bonacich Power Centrality

Pedro Ribeiro – Node Centrality

 Eigenvector Variants
●
There are other variants of eigenvector
centrality, such as:

– PageRank
●
(normalized eigen vector + random jumps)
[we will talk in detail about that later]

– Katz Centrality
●
(connections with distant neighbors are penalized)

Pedro Ribeiro – Node Centrality

 Centrality in Directed Networks
●
Degree:
– in and out centrality
●
Betweenness:
g jk (i)
– Consider only directed paths: C B (i)=∑
j≠k g jk
– When normalizing take care of ordered pairs
' C B (i)
C (i)=
B number of ordered pairs is
(n−1)(n−2) 2x the number of unordered
●
Closeness
– Consider only directed paths
●
Eigenvector (already prepared)
Pedro Ribeiro – Node Centrality
 Centrality in Weighted Networks

●
Degree:
– Sum weights (non-weighted equals weight=1 for all edges)

●
Betweenness and Closeness:
– Consider weighted distance

●
Eigenvector
– Consider weighted adjacency matrix

Pedro Ribeiro – Node Centrality

 Node Centralities: Conclusion
●
There are other node centrality metrics,
but these are the “quintessential”

●
Which one to use depends on what you want to
achieve or measure
– Worry about understanding the concepts
– They enlarge your graph vocabulary
Pedro Ribeiro – Node Centrality
Node Centralities: Conclusion

Betweenness Closeness Eigenvector

Degree Harmonic Katz

Pedro Ribeiro – Node Centrality

 Node Centralities: Conclusion
●
All (major) network analysis packages provide them:

●
Also all (major) network analysis and visualization platforms:

Pedro Ribeiro – Node Centrality