International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1225

A Comparative Study of Page Ranking

Algorithms for Online Digital Libraries

Sumita Gupta, Neelam Duhan, Poonam Bansal

Abstract— With the increasing use of academic digital libraries, it becomes more important for authors to have their publications or scientific literature well ranked in order to reach their audience. Web mining is a potential candidate to meet this challenge. These techniques when applied on the search results of a user query, provide them an order so that users can easily navigate through the search results and find the desired information content. This is also called Page Ranking. The ranking mechanisms can arrange the scientific literature or publications in order of their relevance, importance and content score. In this paper, a survey of some prevalent page ranking algorithms for academic digital libraries is being carried out and comparison of these algorithms in context of performance has been done.

Keywords: Digital libraries, Search engine, Page ranking, Web mining

1 INTRODUCTION

——————————  ——————————
As a consequence of the availability of huge amounts of
digital content, modern search engine technologies are
ith the rapid growth of information sources available on the World Wide Web (WWW) and growing needs of users, it is becoming difficult to manage the information on the Web and satisfy the user needs [1]. For this purpose, many advanced web searching and mining techniques have recently been developed to tackle the problem of finding relevant information and are being used in the commercial search
engines such as Google and Yahoo.
In spite of advances in search engine technologies, there still occur situations where the user is presented
with non-relevant search results. For example, when a
user inputs a query for some scientific literature, book or periodical to a general purpose search engine such as Google, it returns a long list of search results consisting of tutorials, news, articles, blogs etc. This happens due to limited crawling by the search engines. Most of the search engines are not completely capturing the vast amount of information available in the digitization projects on books and periodicals that are occurring locally, nationally and internationally.
Moreover, researchers are making their work available online in the form of postscript or PDF documents, therefore, amount of scientific information and the number of electronic journals on the Web is increasing at a fast rate. But the access to the growing body of scientific literature on the publicly indexable Web is limited by the lack of organization of such information. To overcome this problem, digital libraries have been introduced to make retrieval mechanism more effective and relevant for researchers or users. A digital library [2] is an integrated set of services for capturing, cataloging, storing, searching,
protecting and retrieving information, which provides coherent organization and convenient access to typically large amounts of digital information. Now a day, digital libraries are experiencing rapid growth with respect to both the amount and richness of available digital content.

————————————————

 Sumita Gupta is currently pursuing Ph.D in Computer Engineering from YMCA University of Science & Technology, Faridabad, India, E-mail: sumitagoyal@gmail.com

 Neelam Duhan, Associate Professor, YMCA University Science &

Technology, Faridabad, India

 PoonamBansal, Associate Professor, Maharaja Surajmal Institute of

Technology, Delhi, India

now being introduced in digital libraries to retrieve the relevant content.
The architecture of a typical digital library search system is shown in Fig.1. The main component of this system is a crawler that traverses the hypertext structure in the web, downloads the web pages or harvest the desired papers published in specific venue (e.g. a conference or a journal) and stores them in database. Usually the publications present on WWW are in the form of postscript files or PDF. Therefore, for every user search, a new instance of the agent is created which locates and downloads postscript files identified by “.ps”, “.ps.Z”, or “.ps.gz” extensions. These downloaded files are passed through the document parsing sub agent which extracts the semantic features from the downloaded documents and places them into a database as parsed documents. The parsed documents are routed to an indexing module that builds the index based on the keywords present in the pages.
When the user fires a query in the form of keywords on
the search interface of a digital library search system, it is retrieved by the database search and browsing sub agent which does query processing by taking the user query in proper syntax and returns an HTML formatted response to the user. The search results are usually presented in the form of an ordered list by the application of page ranking algorithms employed by digital libraries.
In this paper, a survey of some prevalent page ranking algorithms for online academic digital libraries has been done and a comparison is carried out. This paper is structured as follows: in Section II, web mining concepts, categories and technologies have been discussed. Section
III provides a detailed overview of some page ranking algorithms with their strengths and weaknesses. Section IV presents an extensive comparison study. Finally in Section V, conclusion is drawn with a light on future suggestions.

2 WEB MINING

Web mining [3, 4] is a means for automatically discovering and explore useful information from the WWW. There are three areas of Web Mining according to the web data used as input. These are Web Content

IJSER © 2013

http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1226

Mining (WCM), Web Usage Mining (WUM), and Web
Structure Mining (WSM).
Web Content Mining (WCM) is a process of scanning and mining the text, pictures and graphs of a Web page to determine the relevance of the content of the web page in accordance to the search query.
Web Structure Mining (WSM) tries to discover the useful knowledge from the structure of hyperlinks catalogs them and generates information such as the similarity and relationship between papers by taking advantage of their hyperlink topology. WSM uses the graph theory to analyze node and connection structure of a web site where web pages act as nodes and hyperlinks as edges connecting two related pages.
The goal of WSM is to generate structured summary about the website and web page. Table 1 gives an overview of the three mining categories [5].

Web Usage Mining (WUM) is a process of identifying the browsing patterns by analyzing the user’s navigational behavior while surfing on the Web. It extracts data stored in server access logs, referrer logs, agent logs, client-side cookies, user profile and Meta data.

3 PAGE RANKING

Today, the main challenge in front of search engines is to efficiently harness scientific work present on the WWW and present relevant results to the user. Web mining techniques are used in order to extract the relevant information and order the documents.

WWW

User

Crawls

Crawling Module

Postscript Files

Query Results

Web Search Interface

Text Converter

Text Extractor

Document Parsing

Sub-Agent

Parsed

Documents

Document Database (Parsed Documents)

Reads

Database Search & Browsing Sub-Agent

Index

Indexing Module

Fig.1. The Architecture of a Digital Library Search System

Table 1. Web Mining Categories

IJSER © 2013 http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1227

To represent the documents in an ordered manner, Page ranking methods are applied which can arrange the documents in order of their relevance and importance. Some of the common page ranking algorithms for online digital libraries have been discussed here as follows.

2.1 Citation Count Algorithm

This is one of the most frequent used ranking
algorithm for measuring a scientist's reputation, and named as Citation Count [6]. This method uses the citation graph of the web to determine the ranking of scientific work. In citation graph, the nodes represent publications, whereas an edge from node i to node j represent a citation from paper i to paper j i.e. a vote from paper i to paper j. This method states that if a publication has more number of citations (incoming links) to it then publication become important. Therefore, it takes backlinks into account to order the publications. Thus, a publication obtains a high rank if the number of its backlinks is high. Citation Count is defined in (1):
CCi = |Ii| (1)
where CCi represents the citation count of publication i,
|Ii | denotes the number of citations (in-degree) of the
publication i.

Example Illustrating Working of CC. To explain the working of Citation Count, let us take an example of citation graph as shown in Fig. 2 and Table 2, where A, B , C, D, E and F are six publications.

 It does not take into account the importance of citing paper i.e. citation from the reputed journal get the equal weightage as the citation from the poor journal.
 If two papers have similar citation count as
publication D and publication F shown in Fig 2, but interestingly publication F is almost 20 years younger than the publication D, thus it had a much smaller time window to accumulate citations. Thus, it does not take into consideration different characteristics of the citations, like their publication date.

2.2 Time dependent Citation Count

Algorithm

Ludmila Marian [7, 8] proposed an extension to standard Citation Count method named Time Dependent Citation Count (TDCC). It is a time-dependent approach which takes into account time of the citation. This method assumes that the freshness of citations and link structure are factors that need to be taken into account in citation analysis while computing the importance of a publication. Thus, Citation Count algorithm is modified by initially distributing random surfers exponentially with age, in favor of more recent publications. The method introduces the effect of time in the citation graph by applying a time-decay factor to the citation counts.
The weight of a publication i is denoted as We‹ghti as
given in (2)

ܹ݁݅݃ℎݐ௜ = ݁ −௪(௧೛ −௧೔ ) (2)

F where ti denotes the published year of publication i, tp

A

denotes the present time (i.e. year), and w denotes the
time decay parameter (w є (0, 1]), which quantifies the

C notions of “new" and “old" citations (i.e. publications

B D with ages less than the time decay parameter would be considered “new"; publications with ages larger than the

E time decay parameter would be considered “old")

citations (in-degree) of the publication i.

Fig. 2 Example of Citation Graph

Table 2. Data of citation Graph

The Citation Count for publications A, B, C, D, E and F can be calculated by using (1):
CC(A)=0, CC(B)=0, CC(C)=3, CC(D)=2, CC(E)=1, CC(F)=2
The ranking of publications based on Citation Count become: CC (C) > (CC (D), CC (F)) > CC (E) > (CC (A), CC (B))

Limitations of CC. There are a number of cases where this method fails to reveal the good picture of influence of publications in its domain. Few of reasons for this are:

Example Illustrating Working of TDCC. To illustrate the working of TDCC, let us refer again to Fig 2 and Table 1. By using (2) weight scores of publications can be calculated as:

WtA = 0 (2a) WtB = 0 (2b) Wtେ = e−w(ଶ଴ଵଶ−ଶ଴ଵଵ) + e−w(ଶ଴ଵଶ−ଶ଴଴8) + e−w(ଶ଴ଵଶ−ଶ଴଴଴)
= e−w(ଵ) + e−w(4) + e−w(ଵଶ) (2c) WtD = e−w(ଶ଴ଵଶ−ଵ998) + e−w(ଶ଴ଵଶ−ଶ଴଴7)
= e−w(ଵ4) + e−w(5) (2d)
WtE = e−w(ଶ଴ଵଶ−ଶ଴଴8)
= e−w(4) (2e)
WtF = e−w(ଶ଴ଵଶ−ଶ଴ଵଵ) + e−w(ଶ଴ଵଶ−ଶ଴଴8)
= e−w(ଵ) + e−w(4) (2f)
where w is time decay factor. Let us take the threshold
age = 6 years i.e. w=0 for the publications with the ages
less than 6 years (considered new publications) and w=1
for publications with ages more than 6 years (considered old publications). By calculating the above equations, the rank score of publications become:

IJSER © 2013 http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1228

TDCC (A) = 0, TDCC (B) = 0, TDCC (C) =2.0000006144
TDCC (D) = 1.000000832, TDCC (E) = 1, TDCC (F) = 2
ܴܲ(ܨ) = (1 െ ݀) + ݀ ቆ

ܴܲ(ܣ)
2 +

ܴܲ(ܤ)
3 ቇ (3݂)
Here TDCC(C)>TDCC (F) > TDCC (D) > TDCC (E) > (TDCC (A), TDCC (B)). It may be noted that the resulting ranking of citations obtained by CC and TDCC is different.

Advantages and Limitations of TDCC. After adding a time decay parameter, the time-dependent ranking can differentiate between an old publication that acquired a large number of citations over a long period of time, and a new publication. That, although important for the scientific community, did not have enough time to acquire as many citations as the old one, in the favor of the latter.

 Adding a week or strong time decay factor to a
ranking method will have an impact on the final ordering of the documents. For example adding a strong time decay factor to ranking will reveal the most popular publications at the current moment in time.
 Like CC, this method does not take into consideration the different importance of each
citation have been discussed here as follows.

2.3 PageRank Algorithm

Surgey Brin and Larry Page [9,10] proposed a ranking
algorithm, named PageRank (PR) which extends the idea
of citation analysis. In citation analysis, the incoming
links are treated as citations which provide importance to
a page but this technique could not provide fruitful
results. In turn, PageRank [10] provides a better approach
which is based on the fact, that the importance of a research paper can be judged by the number of citations the paper has from other research papers. This algorithm states that if a link comes from an important paper then this link is given higher weightage than those which are coming from non-important papers. These links are called as backlinks. The PageRank of a paper u can be calculated as:
PR(v)

Table 3 Iteration Method for PageRank

Let us assume the initial PageRank as 1, d is set to 0.85 and do the calculation. The rank values of papers are iteratively substituted in above page rank equations to find the final values until the page ranks get converged as shown in Table 3.
By calculating the above equations iteratively, the page ranks of papers become:
PR (D) > PR (C) >PR (F) > PR (E) > (PR (A), PR (B))

Advantages and Limitations of PR. One of the main advantages of this method is that it ranks the publications accordingly to the importance of their citations, bringing to light some very insightful publications that would not have been discovered with the Citation Count method. On the other hand, there are some shortcomings of this ranking method also [11]:

 The rank score of publication is equally
distributed among its all references irrespective of assigning the larger rank values to more important papers.
 A page rank of a publication is mostly affected by the scores of the publications that point to it and less by the number of citations. For example, in Fig. 3, node F gets higher score than node E, although node E gets 4 citations and node F gets
1 citation.
 PageRank gives high score to a node u, if it contained a cycle. For Example, Table 4 shows

PR(u) = (1-d) +d ෍
Nv

௩ € ஻(௨)

(3)
the rank results of graph shown in fig 3. In this,
node E gets 4 citations, whereas node T gets 3
citations. However, the PageRank score of node
where u represents a paper, B(u) is the set of papers that point to u, PR (u) and PR (v) are rank scores of
papers u and v respectively, Nv denotes the number of
outgoing links of paper v, and d is a normalization factor.

Example Illustrating Working of PR. Let us take a previous example as shown in Fig 2 in order to explain the working of PageRank algorithm. The PageRanks for

T is about 2 times higher than that of node E. This happens because node T is a part of citation cycle. But in bibliometrics, cycles represents the self-citations which do not occur in citation graph. Thus PageRank does not provide fruitful results in bibliometrics.
papers can be calculated by using (3):
ܴܲ(ܣ) = (1 െ ݀) + ݀(0) (3ܽ)
ܴܲ(ܤ) = (1 െ ݀) + ݀(0) (3ܾ)

A B C D

E

P Q R S

T

ܴܲ(ܥ ) = (1 െ ݀) + ݀ ቆ

ܴܲ(ܣ)
2

ܴܲ(ܤ)
+ 3 +

ܴܲ(ܨ)
1
ቇ (3ܿ)

F

ܴܲ(ܦ) = (1 െ ݀) + ݀ ቆ

ܴܲ(ܥ )
1

ܴܲ(ܧ)
+
1

U

ቇ (3݀)
ܴܲ(ܤ)

ܴܲ(ܧ) = (1 െ ݀) + ݀ ቆ
3

ቇ (3݁)

IJSER © 2013

http://www.ijser.org

Fig. 3. An example of a graph

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1229

Table 4 Rank results Fig. 3

2.4 Popularity Weighted Ranking

Algorithm

Yang Sun and C. Lee Giles [12] gave a new ranking
method based on PageRank with significant
improvement for ranking academic papers, named Popularity Weighted Ranking algorithm. This method combines the concepts that seem to be important for analyzing the importance of publication. The publication importance is determined on the basis of the weighted citations from the other papers and a popularity factor of its publication venue i.e. quality of the publication venue where a publication is published. Unlike impact factor, it does not differentiate between journals, conferences and workshop proceedings. The popularity factor of a publication venue v in a given year is defined by (4)
 This algorithm works well for most queries but it does not work well for others.
 This method assumes that ranking score of a previously published paper will not have any impact on later published ones i.e. it does not take into consideration the time of publication.
 This method also does not differentiate between the popular and prestigious author who published the papers.

2.5 HITS Algorithm

Kleinberg [13] proposed a more refined notion for the
importance of the web pages called Hyperlink Induced
Topic Search (HITS). This method identifies two different
forms of Web pages called hubs and authorities. Authorities are pages having important contents and hubs are pages that act as resource lists, guiding users to authorities as shown in Fig 1. A good authority is a page pointed to by good hubs, while a good hub is a page that points to good authorities. A page may be a good hub and a good authority at the same time.

Hubs Authorities
ܲܨ (ݒ, ݐ) =

nv
× ෍
N

i€ ୔

PF (‹, t) × ™(t) N(t)
(Ͷ)

F Fig 4: Hubs and Authorities i

Input: Root set R; Output: Base set S

where PF (v,t) represents the popularity factor of publication venue v in a given year t, P represents the set
of publication venues i which cite v in that year, nv
denotes the number of papers published in venue v in
that year, w (i) is the weight which represents the
frequency that venue i cites venue v and N (i) denotes the
total number of references generated by venue i. Considering the importance of popularity factor of publication venue, the ranking score of publication p at a previous time t is given in (5).
R(q୲ )

Let S = R

1.For each page p є S, do Steps 3 to 5

2. Let T be the set of all pages S points to.

3. Let F be the set of all pages that point to S.

4. Let S = S + T + some or all of F.

5. Delete all links with the same domain name.

6. Return S

Fig 5: Algorithm to determine Base Set

HITS functions in two major steps.
1. Sampling Step: Sampling Step: In this step, a
ܴ (ݍ் ) = PF(˜୔T ) + ෍

୲>் ,୯౪€D

N(q୲ )

(ͷ)
set of relevant pages for the given query are collected i.e. a sub-graph S of G is retrieved
where R(ݍ௧ ) represents the ranking score of a paper ݍ௧ , which is published at time t and cite paper ݌் , D represents the set of papers which cite ݌் , N(ݍ௧ ) denotes the number of references in paper ݍ௧ , PF(ݒ௣೅ ) denotes the
popularity factor of the publication venue v where paper
݌் is published.

Advantages and Limitations of Popularity Weighted

Ranking Algorithm. One of the main advantages of this method is that it overcomes the limitations of impact factor i.e. by considering the impact of all publication venues and the probability of reader access.

IJSER © 2013

which is high in authority pages [4]. The algorithm starts with a root set R selected from the result list of a digital library search system. Starting with R, a set S is obtained keeping in mind that S is relatively small, rich in relevant pages about the query and contains most of the high authorities. HITS algorithm expands the root set R into a base set S by using the algorithm (see Fig. 5).
2. Iterative Step: This step finds hubs and authorities
using the output of the sampling step. In this [14]
each page is associated with two numbers: an
authority weight ܽ௜ , and a hub weight ℎ௜ . Pages

http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1230

with a higher ܽ௜ value are considered as better authorities and pages with a higher ℎ௜ value as
better hubs.
Let A be the adjacency matrix of the graph S (output of sampling step), v denotes the authority weight vector and
u denotes the hub weight vector.The weights ܽ௜ and ℎ௜ of
all the nodes in S are dynamically updated by as follows:
˜ = (ܣ௧ × u) (6)
u = (ܣ × ˜) (͹)
If we consider that the initial weights of the nodes as
(A), HITS (B))

Limitations of HITS. Following are some constraints of

HITS algorithm [15]:
 Distinction between Hubs and authorities: It is not
easy to distinguish between hubs and authorities because many sites act as hubs as well as authorities.
 Topic drift: Sometime HITS may not produce the
most relevant documents to the user queries because of equivalent weights.
 Automatically generated links: HITS gives equal
importance to automatically generated links
1
⎡1⎤
ݑ଴ = ⎢ . ⎥
⎢ . ⎥
⎣1⎦
1
, then ܣ௧ × ൦1
.
1
which may not have relevance for the user query.

2.6 PaperRank Algorithm

After applying k steps we get the system:
ݒ௞ = (ܣ௧ × ܣ) × ݒ௞−ଵ (6ܽ) ݑ௞ = (ܣ × ܣ௧ ) × ݑ௞−ଵ (͹ܽ)

Example Illustrating Working of HITS. The adjacency matrix of the graph is:

Zhang Guangqian [16] gave a new ranking method for publications ranking named PaperRank based on Google's PageRank. In this method, publication’s rank score is determined on the basis of the reading value and other factor because it considers that the reading value of same papers may be different due to different readers.
0 0 1
⎡0 0 1
⎢
0 0 1
0 1 1⎤
⎥
0 0 0
⎡0 0 0
⎢
0 0 0
0 0 0⎤
⎥
The reading value of a paper is related to its content, the
periodical in which it was published, and the author of the paper. Thus, this method considers the factors such as
ܣ = ⎢0 0 0
1 0 0⎥ A୲ = ⎢1 1 0
0 0 1⎥
content, journal, author, published time etc. in order to
⎢0 0 0
⎢0 0 0
⎣0 0 1
0 0 0⎥
1 0 0⎥
0 0 0⎦
⎢0 0 1
⎢0 1 0
⎣1 1 0
0 1 0⎥
0 0 0⎥
0 0 0⎦
1
⎡1⎤
measure the reading value of papers. PaperRank of the publication p can be calculated as:
ܲܽ݌݁ݎܴܽ݊݇ = ܤܴ × ܣܴ × ܫܨ × ܦ (ͺ)
⎢ ⎥
Assume the initial hub weight vector is: ݑ = ⎢1⎥
⎢1⎥
⎢1⎥
⎣1⎦
We compute the authority weight vector by:
˜ = (ܣ௧ × u)
where BR represents the baserank, AR denotes the AuthorRank, IF denotes the impact factor of the journal in which it was published and D represents the published time of publication p. Various parameters used in the PaperRank calculation are explained below.

BaseRank

0 0 0
⎡0 0 0
⎢
0 0 0
0 0 0⎤
⎥
1
⎡1⎤
0
⎡0⎤
Thes BaseRank (BR) calculates the rank of the publication by using the PageRank algorithm. It considers the quoted
ݒ = ⎢1 1 0
0 0 1⎥ × ⎢1⎥ = ⎢3⎥
time of cited publication and the importance of the citing
⎢0 0 1
⎢0 1 0
⎣1 1 0
0 1 0⎥
0 0 0⎥
0 0 0⎦
⎢1⎥
⎢1⎥
⎣1⎦
⎢2⎥
⎢1⎥
⎣2⎦
publication. The BaseRank formula is given as:

ܤܴ(ݒ) ܤܴ(ݑ) = ܿ ෍
ܰ௩
(ͻ)
Then, the updated hub weight is
u = (ܣ × ˜)

௩€஻(௨)

where u represents a publication, B(u) is the set of
citations that point to u, BR(u) and BR(v) are rank scores
0 0 1
⎡0 0 1
⎢
0 0 1
0 1 1⎤
⎥
0
⎡0⎤
ͷ
⎡6⎤
of publications u and v respectively, ܰ௏ denotes the
number of publications cited by publication v (i.e.
= ⎢0 0 0
1 0 0⎥ × ⎢3⎥ = ⎢2⎥
number of references), c is a factor used for
⎢0 0 0
⎢0 0 0
⎣0 0 1
0 0 0⎥
1 0 0⎥
0 0 0⎦
⎢2⎥
⎢1⎥
⎣2⎦
⎢0⎥
⎢2⎥
⎣3⎦
normalization.

AuthorRank

This parameter assumes that if paper A cited by paper B
By using (6a) and (7a), the authority weights and hub
weights are iteratively calculated untill the values get converged as shown in Table 5.
By calculating the above equations iteratively, the page ranks of papers become:
HITS (C) > HITS (F) >HITS (E) > HITS (D) > (HITS
and C at the same time, then, being cited by paper B authored by a popular and prestigious author contributes more to the Rank value of A than being cited by paper C with an unimportant author. Thus, it calculates the AuthorRank by considering the authors contribution in a certain academic field. The AuthorRank can be computer

IJSER © 2013 http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1231

by using an author citation network [17] which is a directed and weighted graph where nodes represent authors, edges represent citing relationships from author
Thus, it introduces the time factor D as follow:
(ݐ െ ݉݅݊{ܶ(݇)} + 1)
A to author B, and edge weights represent the number of times that author A cites author B. The AuthorRank can
ܦ(݌) =

(݉ܽݔ{ܶ(݇)} + 1) (12)
be calculated as:
ܣܴ(ܽ) = d ෍

௕€஻(ୟ)

ܣܴ(b)
ܰୠ
(10)
where D(p) represents time factor of paper p, t is the year in which p was published, B(p) denotes the set of all the papers, T is a n*1 matrix composed by all the years in which all the papers were published, and n is the total number of all the papers.
where a represents an author , AR(a) is the set of author’s
“citing” author a, AR(a) and AR(b) are AuthorRank of
author’s a and b respectively, ܰ௕ denotes the number of
authors cited by author b, d is a normalization factor.

Impact Factor of Journal

This parameter assumes that if paper A is cited by paper B and C, and paper B was published in the core journal, and paper C was from unimportant journal, then the vote from paper B to A contributes more rank value to paper A than a vote from paper C to paper A. Thus, it considers the impact factor of journal to represent the weight of each journal. The formula for calculating the impact factor of the journal is defined as follow:
ܥ

Limitations of PaperRank. Researchers have shown that scientific publications naturally form a network on the basis of citation relationships. This algorithm can do well for the direct relationships i.e. citation and cited relationships, but it may not adequately reflect the lineage of scientific works. In such scenario, counting the indirect citation, indirect co-citation, and indirect co- reference, which are feasible in the Web environment may be considered.

4 A Comparison Study

By extensive study and literature analysis of some of the

ܫܨ(݆) =
ܣ
(11)
important web page ranking algorithms, it is concluded that
each algorithm has some relative strengths and limitations. A detailed comparison of ranking algorithms studied is
where IF (j) represents the impact factor of journal j, A
denotes the total number of papers published in journal j
in the previous two years, and C denotes the quoted
times of papers in the current year.

Published Time

This parameter considers the time of the publication. It assumes that sometimes a recently published publication having only one or two citations due to small time window may be important to reader in a certain field.
shown in Table 6. Comparison is done on the basis of some measures such as main techniques used, methodology, input parameters, relevancy, quality of results, importance and limitations.

Table 4 Iteration Method for HITS

IJSER © 2013 http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1232

Table 6: COMPARISON OF RANKING ALGORITHMS

*N: NUMBER OF PAPERS, M: AVERAGE CITATIONS TO A PAPER

REFERENCES

[1] Naresh Barsagade, “Web Usage Mining And Pattern

Discovery: A Survey Paper”, CSE 8331,2003

[2] M. Krishnamurthy, “Open access, open source and digital libraries: A current trend in university libraries around the world”, A General Review, Emerald Group Publishing Limited

[3] R. Cooley, B. Mobasher and J. Srivastava, “Web Mining:

Information and Pattern Discovery on the World Wide

Web”, Proceedings of the 9th IEEE International

Conference on Tools with Artificial Intelligence, pp. 558-

567, 1997

[4] N. Duhan, A.K. Sharma and K.K. Bhatia, ”Page Ranking Algorithms: A Survey”, Proceedings of the IEEE International Conference on Advance Computing, 2009

[5] R. Kosala, and H. Blockeel, “Web Mining Research: A Survey”, SIGKDD Explorations, Newsletter of the ACM Special Interest Group on Knowledge Discovery and Data Mining, vol. 2, no. 1, pp. 1-15, 2000

IJSER © 2013 http://www.ijser.org

International Journal of Scientific & Engineering Research, Volume 4, Issue 4, April 2013

ISSN 2229-5518 1233

[6] Joeran Beel, Bela Gipp, “Google Scholar's Ranking Algorithm: The Impact of Citation Counts (An Empirical Study)”, In Rcis 2009: Proceedings of The IEEE International Conference on Research Challenges In Information Science, 2009

[7] L. Marian, M. Rajman, “Ranking Scientific Publications

Based on Their Citation Graph”, Master Thesis, CERN-

THESIS, 2009

[8] L. Marian, J. Yves LeMeur, M. Rajman, M. Vesely, Citation Graph Based Ranking in Invenio”, ECDL, pp.

236-247, 2010

[9] L. Page, S. Brin, R. Motwani, and T. Winograd.P, “The

Pagerank Citation Ranking: Bringing order to the Web. Technical report, Stanford Digital Libraries SIDL-WP-

1999-0120, 1999

[10] Sergey Brin and Larry Page, “The anatomy of a Large-

scale Hypertextual Web Search Engine”, In Proceedings of the Seventh International World Wide Web Conference, 1998

[11] A. Sidiropoulos, Y. Manolopoulos, “Generalized Comparison of Graph-based Ranking Algorithms for Publications and Authors”, Journal of Systems and

Software archive, vol. 79, issue 12, pp. 1679-1700, 2006Y. Sun, C. L. Giles, “Popularity Weighted Ranking for Academic Digital Libraries”, In 29th ECIR, pp. 605-612,

2007

[12] Kleinberg J., ”Authorative Sources in a Hyperlinked Environment”, Proceedings of the 23rd annual International ACM SIGIR Conference on Research and Development in Information Retrieval, 1998

[13] Math Explorer's Club, “The Mathematics of Web Search”, http://www.math.cornell.edu/~mec/Winter2009/Ralu caRemus/Lecture4/lecture4.html

[14] Longzhuang Li, Yi Shang, Wei Zhang, “ Improvement of HITS based Algorithms on Web Documents”, WWW2002,Honolulu Hawaii USA, 2002

[15] Zhang Guangqian, Liu Xin, “Study on the Method of

Ranking Scientific Papers”, In The International Conference on E-Business and E-Government, Guangzhou China, pp. 3062-3066, 2010

[16] Ding Y., “Applying Weighted PageRank to Author Citation Networks”, Journal of the American Society for Information Science and Technology, pp. 236-245,

IJSER © 2013 http://www.ijser.org