PageRank

Algorithm

Prepared By: Mai Mustafa

Contents:

Background
Introduction to PageRank
PageRank Algorithm
Power iteration method
Examples using PageRank and iteration
Exercises
Pseudo code of PageRank algorithm
Searching with PageRank
Application using PageRank
Advantages and disadvantages of PageRank algorithm
References

Background
PageRank was presented and published by
Sergey Brin and Larry Page at the Seventh
International World Wide Web Conference
(WWW7) in April 1998.
The aim of this algorithm is track some
difficulties with the content-based ranking
algorithms of early search engines which used
text documents for webpages to retrieve the
information with no explicit relationship of link
between them.

Introduction to PageRank
PageRank is an algorithm uses to measure
the importance of website pages using
hyperlinks between pages.
Some hyperlinks point to pages to the same
site (in link) and others point to pages in
other Web sites(out link).
PageRank is a vote, by all the other pages
on the Web, about how important a page is.
A link to a page counts as a vote of support

PageRank Algorithm
The main concepts:
In-links of page i : These are the hyperlinks that point to page i
from other pages. Usually, hyperlinks from the same site are not
considered.
Out-links of page i : These are the hyperlinks that point out to
other pages from page i .

The following ideas based on rank prestige are used

to derive the PageRank algorithm:
A hyperlink from a page pointing to another page is an implicit
conveyance of authority to the target page. Thus, the more inlinks that a page i receives, the more prestige the page i has.
Pages that point to page i also have their own prestige scores. A
page with a higher prestige score pointing to i is more important
than a page with a lower prestige score pointing to i .

Cont. PageRank Algorithm

To formulate the above ideas, we treat the Web as
a directed graph G = (V, E), where V is the set of vertices
or nodes, i.e., the set of all pages, and E is the set of
directed edges in the graph, i.e., hyperlinks. Let the
total number of pages on the Web be n (i.e., n = |V|).

The PageRank score of the page i (denoted by P(i)) is

defined by:
1-

Oj is the number of out-links of page j.

Cont. PageRank Algorithm

Mathematically, we have a system of n linear equations
(1) with n unknowns. We can use a matrix to represent
all the equations. Let P be a n-dimensional column vector
of PageRank values

Let A be the adjacency matrix of our graph with

2-

We can write the system of n equations with

3-

Cont. PageRank Algorithm

the above three conditions come from Markov chains Model,
in it; each Web page in the Web graph is regarded as a
state. A hyperlink is a transition, which leads from one state
to another state with a probability. Thus, this framework
models Web surfing as a stochastic process. It models a
Web surfer randomly surfing the Web as a state transition in
the Markov chain .so on, this three conditions are not
satisfied. Because First of all, A is not a stochastic matrix. A
stochastic matrix is the transition matrix for a finite Markov
chain whose entries in each row are nonnegative real
numbers and sum to 1. This requires that every Web page
must have at least one out-link. This is not true on the Web
because many pages have no out-links, which are reflected
in transition matrix A by some rows of complete 0s. Such
pages are called dangling pages (nodes).

Cont. PageRank Algorithm

We can see that A is not a
stochastic matrix because the fifth
row is all 0s, that is, page 5 is a
dangling page.
We can fix this problem by adding
a complete set of outgoing links
from each such page i to all the
pages on the Web. Thus, the
transition probability of going from
i to every page is 1/n, assuming a
uniform probability distribution.
That is, we replace each row
containing all 0s with e/n, where e
is n-dimensional vector of all 1s.

Cont. PageRank Algorithm

Another problems:
A is not irreducible, which means that the Web graph G is not
strongly connected. And to be strongly connected it must have
a path from u to v. (if there is a non-zero probability of transitioning
from any state to any other state).
A is not aperiodic. A state i in a Markov chain being periodic means
that there exists a directed cycle that the chain has to traverse. To
be aperiodic all paths leading from state i back to state i have a
length that is a multiple of k.
It is easy to deal with the above two problems with a single
strategy. We add a link from each page to every page and give each
link a small transition probability controlled by a parameter d, it is
used to model the probability that at each page the surfer will
become unhappy with the links and request another random page.
The parameter d, called the damping factor, can be set to a value
between 0 and 1.
Always d = 0.85.

Cont. PageRank Algorithm

The PageRank model:
AT (t: Transpose)

The PageRank formula for each page i :

Power iteration method:

The PageRank algorithm must be able to deal with billions of pages,
meaning incredibly immense matrices; thus, we need to find an efficient
way to calculate the eigenvector of a square matrix with a dimension in the
billions. Thus, the best option for calculating the eigenvector is through the
power method. The power method is a simple and easy to implement
algorithm. Additionally, it is effective in that it is not necessary to compute a
matrix decomposition, which is near-impossible for matrices containing very
few values, such as the link matrix we receive. The power method does have
downsides, however, in that it is only able to find the eigenvector of the
largest absolute-value eigenvalue of a matrix. Also, the power method must
be repeated many times until it converges, which can occur slowly.
Fortunately, as we are working with a stochastic matrix, the largest
eigenvalue is guaranteed to be 1. Since this is the eigenvector we are
searching for, the power method will return the importance vector we are
looking for. Additionally, it has been proven that the speed of convergence
for the Google PageRank matrix is slower the closer gets to 0. Since we
have set d to be equal to 0.15, we can expect the speed of convergence to
be approximately 50 - 100 iterations, which is the number of iterations
reported by the creators of PageRank to be necessary for returning
sufficiently close values.

Simple example using PageRank with

iteration
2 pages A,B:
P(A)=(1-d)+d(pagerank(B)/1)
P(A)=0.15+0.85*1=1

P(B)=(1-d)+d(pagerank(A)/1)
P(B)=0.15+0.85*1=1

When we calculate the PageRank of A and B is 1. now, we

plug in 0 as the guess and calculate again:
P(A)=0.15+0.85*0=0.15
P(B)=0.15+0.85*0.15=0.2775
Continue the second iteration:
P(A)=0.15+0.85*0.2775=0.3859
P(B)=0.15+0.85*0.3859=0.4780
If we repeat the calculations, eventually the PageRank for both
the pages converge to 1.

Another example using PageRank with

iteration
Three pages A,B And C
P(A)=(1-d)+d(pagerank(B)+pagerank(C)/1)
P(B)=(1-d)+d(pagerank(A)/2)
P(C)=(1-d)+d(pagerank(A)/2)
Begin with the initial value as 0:
1st iteration:
P(A)=0.15+0.85*0=0.15
P(B)=0.15+0.85*(0.15/2)=0.21
P(c)=0.15+0.85*(0.15/2)=0.21
2nd iteration:
P(A)=0.15+0.85*(0.21*2)=0.51
P(B)=0.15+0.85*(0.51/2)=0.37
P(C)=0.15+0.85*(0.51/2)=0.37

Cont. example
3rd iteration:
P(A)=0.15+0.85*(0.37*2)=0.78
P(B)=0.15+0.85*(0.87/2)=0.48
P(C)=0.15+0.85*(0.87/2)=0.48

And so on.. After 20 iterations

P(A)=1.46
P(B)=0.77
P(C)=0.77
The total PageRank =3, but we can see A has much
larger proportion of the PageRank than B and C, because
they are passing to A not to any other pages.

Exercise:
Given A below, obtain P by solving Equation PageRank model
directly.

first: we will represent the matrix

as graph:

p (1 d )e dA p
T

Find AT

0
1

3
1
3
T
A
1

3
0

first

then find e:

1
2

1
2

0
0

0
0

1
4
1
4

1
2
1
2

1
4
1
4

0
0

2
1
2
0

1
6
1

6
1
6
e
1

6
1
6
1

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1

6
1
6
1

6
1
6
1

And we know that d=0.85

1
6
1

6
1

p 0.15 6
1

6
1
6
1

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1
6
1
6
1
6
1
6
1
6

1
6
1
6
1
6
1
6
1
6
1
6

6 0.85

0.025 0.025 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.025 0.025 0.025

0.283
0

0.025 0.025 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.025 0.025 0.025

1
2

1
2

0
1
3
1
3
1
3

1
4
1
4

1
2
1
2

1
4
1
4

0
0
0

0
0
0

0
0.2125
0.2125

2
1
2

0
0.425
0
0
0
0.283
0
0.85 0.2125 0.425

0.283 0.425
0
0.2125 0.425

0
0
0

0
0
0

0.425
0.425

 0.025
0.308

0.308
p
0.308
0.025

0.025

0.45
0.025
0.45
0.025
0.025
0.025

0.025
0.875
0.025
0.025
0.025
0.025

0.025
0.2375
0.2375
0.025
0.2375
0.2375

0.025
0.45
0.45
0.025
0.025
0.025

0.025
0.025
0.025

0.45
0.45

0.025

Exercise2:
Given A as in problem 1 in the last exercise, use the power
iteration method to show the first 5 iterations of P.
First iteration:
0.025 0.45 0.025 0.025 0.025 0.025
0.308 0.025 0.875 0.2375 0.45 0.025

0.308

0.025

0.308 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.2375 0.025

0.45

k0

0.45

0.45

0.025 0.2375

0.45

0.025 0.025 0.025 0.2375 0.025 0.025

1
0.575
1
1.921

1
1.496

1
0.858

1
0.788

1
0.363

Second iteration:
0.025 0.45 0.025 0.025 0.025 0.025 0.575
0.966
0.308 0.025 0.875 0.2375 0.45 0.025 1.921
2.102

0.308 0.45 0.025 0.2375 0.45 0.025 1.496

1.647
k1 P * k 0

0
.
308
0
.
025
0
.
025
0
.
025
0
.
025
0
.
45
0.858
0.467

0.025 0.025 0.025 0.2375 0.025 0.45 0.788

0.487

0.025 0.025 0.025 0.2375 0.025 0.025 0.363

0.333

k0

 third iteration:
0.025 0.45 0.025 0.025 0.025 0.025
0.308 0.025 0.875 0.2375 0.45 0.025

0.308 0.45 0.025 0.2375 0.45 0.025

0.966
2.102

0.308 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.2375 0.025

0.45

0.467
0.487

0.565
0.391

0.025 0.025 0.025 0.2375 0.025 0.025

0.333

0.250

k 2 P * k1

1.043
2.130

1.647

1.623

0.45

Fourth iteration:
1.043
2.130

0.308 0.025 0.025 0.025 0.025

0.025 0.025 0.025 0.2375 0.025

0.45

0.565
0.391

0.551
0.377

0.025 0.025 0.025 0.2375 0.025 0.025

0.250

0.270

k3 P * k 2

0.025 0.45 0.025 0.025 0.025 0.025

0.308 0.025 0.875 0.2375 0.45 0.025

0.308 0.45 0.025 0.2375 0.45 0.025

1.055
2.111

1.623

1.637

0.45

 Fifth iteration:
0.025 0.45 0.025 0.025 0.025 0.025 1.055
1.047
0.308 0.025 0.875 0.2375 0.45 0.025 2.111
2.118

0.308 0.45 0.025 0.2375 0.45 0.025 1.637

1.623
k 4 P * k3

0
.
308
0
.
025
0
.
025
0
.
025
0
.
025
0
.
45
0.551
0.563

0.025 0.025 0.025 0.2375 0.025 0.45 0.377

0.382

0
.
025
0
.
025
0
.
025
0
.
2375
0
.
025
0
.
025
0.270
0.267

We would then continue this iterating until the values are

approximately stable, and we would be able to determine the
importance ranking using the resulting vector. With this, we
can see that even with a small count of 5 iterations, our vector
was already converging towards the eigenvector. Since this is
the importance vector of our network, we can see that the
PageRank importance ranking of our pages would thus be

2>3>1>4>5>6

Pseudo code of PageRank

algorithm:

Searching with PageRank

Two search engines:

Title-based search engine

Full text search engine

Title-based search engine

Searches only the Titles

Finds all the web pages whose titles contain all the query words
Sorts the results by PageRank
Very simple and cheap to implement
Title match ensures high precision, and PageRank ensures high
quality

Full text search engine

Called Google
Examines all the words in every stored document and also
performs PageRank (Rank Merging)
More precise but more complicated

Cont. searching with

PageRank

Application using
PageRank
the first and most obvious application of the PageRank algorithm
is for search engines. As it was developed specifically by Google
for use in their search engine, PageRank is able to rank websites
in order to provide more relevant search results faster.
applied PageRank algorithm is towards searching networks
outside of the internet. this can be applied towards academic
papers; by using citations as a substitute for links, PageRank can
determine the most effective and referenced papers in an
academic area.
real-world application of the PageRank algorithm; for example,
determining key species in an ecology. By mapping the
relationships between species in an ecosystem, applying the
PageRank algorithm allows the user to identify the most important
species. Thus, being able to assign importance towards key
animal and plant species in an ecosystem allows for easier
forecasting of consequences such as extinction or removal of a
species from the ecosystem.

Advantage and disadvantages of

PageRank algorithm:
Advantages of PageRank:
1. The algorithm is robust against Spam since its not
easy for a webpage owner to add in links to his/her
page from other important pages.
2. PageRank is a global measure and is query
independent.
Disadvantages of PageRank:
3. it favors the older pages, because a new page, even
a very good one will not have many links unless it is
a part of an existing site.
4. It is very efficient to raise your own PageRank, is
buying a link on a page with high PageRank.

References:
Comparative Analysis Of Pagerank And
HITS Algorithms, by: Ritika Wason.
Published in IJERT, October - 2012.
The top ten algorithms in data mining,
by: Xindong wu and vipin kumar.
Building an Intelligent Web: Theory and
Practice, By Pawan Lingras, Saint Mary.
Hyperlink based search algorithmsPageRank and HITS, by: Shatakirti.