Scribd
Upload
18 views20 pages

Graph Neural Networks (S2-23 - AIMLCZG514) : Pilani Pilani

Uploaded by

Mukesh Saini
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
18 views20 pages

Graph Neural Networks (S2-23 - AIMLCZG514) : Pilani Pilani

Uploaded by

Mukesh Saini
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 20

Graph Neural Networks

(S2-23_AIMLCZG514)
BITS Pilani
Pilani Campus
BITS Pilani
Pilani Campus

Module II
Prerequisite

Degree and Degree distribution

BITS Pilani, Pilani Campus


Dynamics of Graphs

 Graphs which arise in many large scale applications


such as the web and social networks satisfy many
properties such as
Degree Distribution
The power law distribution
Sparsity, and Small Diameters
Densification, and Shrinking Diameters

BITS Pilani, Pilani Campus


Why to study Structure of graphs

 What does the web look like?

NO
 How about a social network, like LinkedIn or Facebook?
 Is it normal for a person to have 500 “friends?”
 Is it normal to contact all 500 with equal frequency?
 Which are the most important nodes in a graph?
 So the main task is to identify:
 What normal is (no proper definition)?
Based on that we can spot anomalies (strange/abnormal
behaviour) !
Informally, “normal” behaviour is the pattern that the
majority follows, and “abnormal” behaviour is the deviation.
 How to define importance?
??
BITS Pilani, Pilani Campus
Why to study Structure of graphs

The graph encompasses a large number of applications:


 Social networks
 Facebook and LinkedIn
 Who-Calls-Whom in a phone network,
 Who-Texts-Whom, or
 Who-Collaborates with Whom.
 Cyber-security (for computer networks)
 The goals are to spot strange behaviours, like botnets, port-scanners,
exfiltrators
 Web and Ranking
 Accelerate or control information/ rumours flow.
 influence propagation
 e-commerce
 Amazon, Netflix, Ebay, (rating recommendation, ...

….
BITS Pilani, Pilani Campus
several reasons to study patterns
and laws in real graphs:
 Understanding human (machine, protein) behaviour:
Discovering patterns of connectivity to develop
domain expertise
 Anomaly detection:
Extremely short/long diameter wrt six degree of
connectivity would be suspicious
 Extrapolations:
Prediction or Forecasting
 Generator design:
Real graphs often have privacy issues and/or high
monetary value, and cannot be released

BITS Pilani, Pilani Campus


several reasons to study patterns
and laws in real graphs:
 Graph Compression:
Web-crawls/Facebook with billions of pages/active
users”
While writing, keeping (and compressing)
historical data is important.
Older mails in your mailbox

BITS Pilani, Pilani Campus


BITS Pilani
Pilani Campus

Distance, Degree Distribution,


Radius & Diameter
Distance between two vertices

Distance is a metric space:

• d(u,v)>= 0 and d(u,v) = 0 iff u=v;


• d(u,v) = d(v,u) A
G
I

B
• d(u,w) + d(w,v) >=d(u,v)
H

C
F J

D
E

BITS Pilani, Pilani Campus


Degree Distribution

• Degree Distribution
I
• Eccentricity, A
G

• Diameter & B

• Radius Solution.xlsx
H

C
F J

BITS Pilani, Pilani Campus


Degree Distribution A
B
G
I

• Degree Distribution C
F J
• Eccentricity, D
E
• Diameter &
• Radius Solution.xlsx
I
A G
B

C
F J

D
E

BITS Pilani, Pilani Campus


DIAMETER

 Diameter (for static graph):


The maximum distance between any two nodes,
where distance is the minimum number of hops (i.e.,
edges that must be traversed) on the path from one
node to another, usually ignoring directionality.
One liner “maximum-length shortest path between
all possible pairs”
Intuitively, the diameter represents how much of a
“small world” the graph is – how quickly one can get
from one “end” of the graph to another.
*ANF [1]: works O((N + E)d) time and O(N) space

[1] Christopher Palmer, Phil B. Gibbons, and Christos Faloutsos. ANF: A fast and scalable
tool for data mining in massive graphs. 2002. ACM Press. BITS Pilani, Pilani Campus
DIAMETER

 Diameter (for static graph):


 Hop-plot (for static graph):
Starting from a node u in the graph, we find the
number of nodes Nh(u) in a neighbourhood of h hops.
We repeat this starting from each node in the graph,
and sum the results to find the total neighbourhood
size Nh for h hops ().
The hop-plot is just the plot of Nh versus h.

BITS Pilani, Pilani Campus


DIAMETER

 Diameter (for static graph):


 Hop-plot (for static graph):
 Effective diameter or eccentricity (for static graph):
This is the minimum number of hops in which some
fraction (typically 90%) of all connected pairs of nodes
can reach each other.

BITS Pilani, Pilani Campus


DIAMETER

 Diameter (for static graph):


 Hop-plot (for static graph):
 Effective diameter or eccentricity (for static graph):
 Characteristic path length
For each node in the graph, consider the shortest
paths from it to every other node in the graph.
Take the average length of all these paths.
Now, consider the average path lengths for all
possible starting nodes, and take their median.

BITS Pilani, Pilani Campus


DIAMETER

 Diameter (for static graph):


 Hop-plot (for static graph):
 Effective diameter or eccentricity (for static graph):
 Characteristic path length
 Average diameter:
This is calculated in the same way as the
characteristic path length, except that we take the
mean of the average shortest path lengths over all
nodes, instead of the median.

BITS Pilani, Pilani Campus


DIAMETER

 The advantage of eccentricity is :


its definition works, as is, even for disconnected
graphs, whereas we must consider only the largest
component for the characteristic and average
diameters.
Characteristic path length and eccentricity are less
vulnerable to outliers than average diameter, but
average diameter might be the better if we want worst
case analysis.

BITS Pilani, Pilani Campus


DIAMETER (Examples)

 The diameters of several naturally occurring graphs have


been calculated, and in almost all cases they are very
small compared to the graph size.
Type of Network Diameter
Internet AS level graph (Autonomous system ) 4
Router level graph 12
97%-effective diameter for the Internet Router graph 15
WWW, when a path exists at all)- directed 16
WWW, when a path exists at all)- undirected 7
Facebook social Network 4
Yahoo web 7
instant-messenger (IM) network in Microsoft 7
network of actors 3.65

BITS Pilani, Pilani Campus


BITS Pilani
Pilani Campus

Thank You

You might also like