A DOMAIN-RELATED AUTHORITY MODEL FOR WEB PAGES
BASED ON SOURCE AND RELATED INFORMATION
Liu Yang, Chunping Li and Ming Gu
Key Laboratory for Information System Security, Ministry of Education
Tsinghua National Laboratory for Information Science and Technology, School of Software
Tsinghua University, Beijing, China
Keywords: Authority Model, Link Analysis, Source Information, Related Information, Finance.
Abstract: The Internet has become a great source for searching and acquiring information, while the authority of the
resources is difficult to evaluate. In this paper we propose a domain-related authority model which aims to
calculate the authority of web pages in a specific domain using the source and related information. These
two factors, together with link structure, are what we mainly consider in our model. We also add the domain
knowledge to adapt to the characteristics of the domain. Experiments on the finance domain show that our
model is able to provide good authority scores and ranks for web pages and is helpful for people to better
understand the pages.
1 INTRODUCTION
The Internet is playing an important role in our daily
lives. People now can easily search and acquire
almost everything on the web, for example, news,
movies and pictures. With the numerous information
and resources we are provided by the Internet, it is
convenient to learn better about the world around us.
However, sometimes we may find it difficult to
judge the importance and authority of web pages.
Among millions of web pages the users obtain from
the Internet, it is easy to get lost.
PageRank (Brin and Page, 1998) and HITS
(Kleinberg, 1999) are two algorithms which both
provide good ways to evaluate web pages using link
analysis. Following the idea of these two algorithms,
there are some other algorithms proposed
afterwards, and most of them also apply the idea of
link analysis, e.g. PHITS (Cohn and Chang, 2000),
and SALSA (Lempel and Morgan, 2001). With the
help of these algorithms, people will know better
about the importance of web pages.
However, in some practical situations, more
specific and targeted evaluations on web pages
would be better. The rankings of the traditional
evaluations are general, and not limited to a specific
domain. But pages considered important in some
subject domains may not be considered important in
others (Bharat and Mihaila, 2001). If the evaluation
of web pages adapts to the domain which the pages
belong to, it would be much better for users. For
example, in finance domain, the web pages mostly
have the characteristics of the domain, and their
readers are limited to a certain group. Therefore they
may not have good scores calculated by PageRank
and HITS. But within the related domain, they may
be important and authoritative. Since the scores
given by normal link analysis are not appropriate for
these pages, we should find out another way.
The goal of our research is to design and
implement a domain-related authority model for web
pages which is able to evaluate their authorities. Our
authority model is based on link analysis and other
characteristics of web pages, i.e. the source and
related information, and will be applied to a specific
domain. It will help the users to decide whether a
web page is trustworthy. Because of the variety and
complexity of web pages and the particularities of
different domains, it is necessary to analyze the
characteristics of the Internet and design a model
which meets the domain-specific requirements.
The rest of the paper is organized as follows. In
Section 2 we introduce some related work on the
ranking of web pages. Our authority model based on
the source and related information is presented in
Section 3, followed by experiments and evaluations
245
Yang L., Li C. and Gu M. (2010).
A DOMAIN-RELATED AUTHORITY MODEL FOR WEB PAGES BASED ON SOURCE AND RELATED INFORMATION.
In Proceedings of the 5th International Conference on Software and Data Technologies, pages 245-253
DOI: 10.5220/0002998702450253
Copyright
c
 SciTePress

in finance domain described in Section 4. Finally we
make our conclusions in Section 5.
2 RELATED WORK
Link analysis is introduced by PageRank and HITS,
where hyperlink structures are used to determine the
relative authority of web pages and produce
improved algorithms for the ranking of web search
results (Borodin et al., 2005). The existing
algorithms can be divided into three classes, i.e. the
algorithms based on Random Walk, the hub and
authority framework and the probabilistic model.
Since first proposed in 1998, PageRank has been
improved in these years. Eiron et al. (2004) refine
the basic paradigm to take into account several
evolving prominent features of the web, and propose
several algorithmic innovations. The mathematical
analysis of PageRank when the damping factor α
changes is given in (Boldi et al., 2005), and an
approach to accelerate the iterating computation of
PageRank is proposed in (McSherry, 2005).
The hub and authority framework is proposed by
Kleinberg in HITS, and used a lot in other similar
algorithms, in which the framework is improved and
combined with other information. For instance, in
(Borodin et al., 2001) the authors introduce the
Hub-Averaging-Kleinberg, Threshold-Kleinberg,
and Breadth-First-Search based on the framework.
PHITS (Cohn and Chang, 2000) is a statistical
hubs and authorities algorithm, and a joint
probabilistic model of document content and
hyperlink connectivity is suggested by Cohn and
Hofmann (2000). An alternative algorithm, SALSA
(Lempel and Morgan, 2001) combines ideas from
both HITS and PageRank. A Bayesian algorithm is
also introduced in (Borodin et al., 2001).
All the algorithms above use hyperlink structures
to calculate the authority of web pages, and have
good experiment results. Brian Amento points out in
(Amento et al., 2000) that the result of link analysis
algorithms is consistent with that of human experts,
and there are no significant differences between
different types of link analysis algorithms. However,
none of the algorithms is perfect for all kinds of
situations. In experiments people find out that
different algorithms emerge as the “best” for
different queries, while there are queries for which
no algorithm seems to perform well (Borodin et al.,
2001).
TruthFinder (Yin et al., 2008) is another
algorithm which studies how to find true facts from
a large amount of conflicting information on many
subjects that is provided by various websites. This
algorithm utilizes the relationship between websites
and their information, and finds true facts among
conflicting information and identifies trustworthy
websites better than the popular search engines. The
idea of TruthFinder is similar to link analysis
algorithms, but the goal of the algorithm is different.
The algorithms mentioned above all provide the
general evaluations on web pages, which are good
for general requirements. Besides that, there are
some topic-based algorithms coming up in recent
years, whose rankings are more specific than those
of the traditional algorithms. Topic-Sensitive
PageRank (Haveliwala, 2002) is evolved from the
traditional PageRank. It calculates a vector for every
web page based on several topics. A score vector for
each page is also applied in (Nie et al., 2006) to
distinguish the contribution from different topics,
using a random walk model that probabilistically
combines page topic distribution and link structure.
In these algorithms the score of a web page is not a
single value, but a vector with regard to different
topics. These researches consider the contents of
web pages and provide the rankings on different
topics. With the result of these researches, people
are capable of knowing the importance of web pages
in different areas. However, the rankings rely on the
partition of topics in the algorithms. In our authority
model, we aim to provide the domain-related
ranking. To achieve this, the domain knowledge is
added to our model, instead of partitioning the web
pages into several topics. In this way, our authority
model is able to be applied to different domains and
provide the specific and targeted evaluations.
3 THE AUTHORITY MODEL
FOR WEB PAGES
With the development of the Internet, people are
provided with various information and resources. It
is quite easy for us to acquire the information on the
web. However, due to the variety of web pages, it is
difficult for people to judge the quality of web pages,
and decide whether to trust what the pages say. Link
analysis is able to give the authority score for every
web page. Following the idea of this technique, we
can design and implement an authority model which
aims to calculate the authority of web pages in a
specific domain. In this way, users will know better
about the trustworthiness of web pages when they
browse them.
There are two aspects which are taken into
ICSOFT 2010 - 5th International Conference on Software and Data Technologies
246

account, the source information and related pages,
which will be combined with link structure in our
model, as shown in Figure 1. These two aspects are
analyzed and extracted from the contents of web
pages, and able to reflect the authority of web pages.
The idea of our model is to calculate the authority of
web pages by combining the information extracted
from the contents with link analysis. In this way, our
model will be more reasonable and effective. Details
are described in the following sections.
Figure 1: Aspects considered in our model.
3.1 Source Information
The authority of web pages partly lies in the source
that releases the information. When we read web
pages or newspapers, we will usually notice who the
author is and judge the quality of the resource
accordingly. Therefore, the source information is
important to our authority model, and it accords with
people’s intuition.
The source information can often be found in the
content of the web page, especially for news pages,
which is denoted in Figure 2. Usually most of news
web pages provide the name and the website address
of their sources, which is convenient for our model
to automatically extract them from the contents of
web pages. For those pages whose sources are
unable to extract, their authorities should be low
from a human perspective. Therefore source
information is not considered for these pages during
the calculation of our model.
After extracting the source information from web
pages, we need to find a way to obtain the
importance score for every source. In this paper, we
employ the traffic rank provided by Alexa
(http://www.alexa.com/) as the importance scores
for sources. Alexa is a website which provides the
rankings for websites all over the world. The
rankings in Alexa are based on the number of visits
to websites, which are convenient and reasonable for
source ranking. As a matter of fact, the reason why
we choose Alexa is that the ranking data of websites
is easy to automatically obtain from it and that its
rankings are able to describe the importance of
sources. In addition to the source importance
obtained from Alexa, we also add domain
knowledge to make our source rankings more
appropriate for the domain, which will be described
in Section 4.1.
Figure 2: Source information in a web page.
3.2 Related Web Pages
There are relationships existing among web pages.
In our authority model, the authority of a web page
is not only decided by itself and its source, but also
influenced by related pages. Hyperlinks are
considered as the relationships among pages in link
analysis, but they are not enough for our model.
The relationship we take into consideration in
our model is about the relativity of web pages.
Nowadays most of the news web pages provide the
related information through hyperlinks in the page,
as shown in Figure 3. These lists of relative
hyperlinks are organized by human editors, which
list the topics and news related to the current page.
This is reliable for the authority model, since the
information is picked up and categorized by editors
who are familiar with the news and its background.
Therefore, it is helpful for computing the authority
of web pages. Also, the related information is
extracted from a specific section of the contents
based on the properties of these pages, rather than all
the hyperlinks in the pages. Therefore, it is related
more closely to the contents of the pages, which is
more reasonable for the calculation of authority and
explains the difference between related pages and
normal hyperlinks.
By automatically extracting the related
information from web pages, we can utilize it in our
A DOMAIN-RELATED AUTHORITY MODEL FOR WEB PAGES BASED ON SOURCE AND RELATED
INFORMATION
247

model and calculate the authorities of web pages
with the useful knowledge.
Figure 3: The related information in a web page from
Yahoo!.
3.3 The Authority Model
There are two objects considered in our model, web
pages and sources. The calculation of the authority
model is on the basis of the relationships between
sources and web pages, and between different pages.
The relationships between sources and web pages
are extracted from the contents of the pages, which
describe the organizations which publish the articles.
Hyperlinks are considered as the relationships
between pages, and related information extracted
from pages helps to establish the relativity of web
pages.
In our authority model, link structure, source
information and related pages are combined together
to compute the authorities of web pages. The
calculation is performed in two steps. In Step 1 we
utilize the link structure in an iteration process. At
first the authorities of all pages are initialized to 1.
After that, the authorities are calculated iteratively.
During each iteration, the authority of a web page is
updated with the sum of the authorities of the web
pages which point to it and are pointed to by it. The
corresponding hyperlinks are named as in-links and
out-links in the following sections. Normalization is
done after each iteration. Considering that the web
pages which point to the current page should have
more influences on its authority than those pointed
to by it, we set up two parameters, α and β, to adjust
the weights of these two kinds of links. The iteration
stops when the authorities of all web pages
converge. The calculation in each iteration is shown
in Equation (1).
a

p

α∑

aq  β  ∑

ar
(1)
In Equation (1) p, q and r denote a web page
respectively, a(p) represents the authority of web
page p, α and β are the adjusting parameters of
in-links and out-links, and 0 < β < α ≤ 1. qp
means the web page q has a hyperlink which points
to p.
During the iteration process, the authorities of
web pages are computed based on the link structure.
In order to gain more appropriate and reasonable
scores for web pages, we need to add the source and
related information. Therefore in Step 2 the
importance of the source and the authorities of
related pages are added to the authority of the web
page. The weights of source and related pages are
adjustable by changing the values of the parameters.
a

p

aa

p

bs

p

c
∑
RP
ar
(2)
In the above equation, s(p) is the importance of the
source of page p, RP(p) represents the set of related
pages for web page p, and a, b, c are the adjusting
parameters for the three elements in Equation (2), a
+ b + c = 1. How to select the values of the
parameters in Equation (1) and (2) will be presented
in Section 4.
In Equation (2) the source importance and the
authorities of related pages are added to the original
authority with different weights. In this way, the two
factors which influence the authority of a web page
are combined into our model.
There are three factors we take into consideration
in our model, link structure, source importance and
related pages. Using link structure is the idea of
traditional algorithms, while source importance and
related pages are the information extracted from the
contents of web pages. They are all important for the
authorities of web pages. Calculating only one of
them is insufficient. Therefore we combine them
together to form a complete evaluation of web pages.
With this model, we are able to precisely evaluate
the authorities of web pages, and help the users to
make a better judgment.
4 EXPERIMENTS
AND EVALUATIONS
The effectiveness of our authority model is
evaluated on the finance domain. The web pages in
ICSOFT 2010 - 5th International Conference on Software and Data Technologies
248

the domain are about financial news and comments,
and their contents are professional and limited to the
specific field. In order to adapt to the characteristics
of finance domain, we need to add the domain
knowledge to our model. Besides that, the
calculation of web page authorities is the same as
introduced in Section 3.
Our experiment is based on the web pages
crawled from the Internet. Pages need to be
processed after crawling to extract the necessary
information for our experiment. Our authority model
is applied to the web pages after that. In order to
better evaluate the experiment result, we use a
method to partition the authorities of web pages into
different ranks, and a manually annotated set is used
for evaluation. The detailed description and analysis
are presented below.
4.1 Adding the Domain Knowledge
In our experiment, the authority model is applied to
the finance domain. Therefore, the domain
knowledge is quite necessary to judge the authority
of web pages. The method of adding the domain
knowledge to our model is mainly to adjust the
importance of sources according to the features of
the domain. In Section 3.1 we introduced our
method of getting importance scores from Alexa,
which are the general rankings on the basis of daily
visits to websites. However, the area of finance has
its own characteristics, which cannot be obtained
simply from Alexa. For example, China Stock is a
famous and professional website on finance in
China, but its importance in Alexa is not ranked
highly. Due to the specialty of financial websites and
their limits of scopes, the websites usually do not
have many visits, and their visitors are people who
are interested in finance and have the background
knowledge, rather than the normal Internet users.
Hence, we may find those professional websites to
be ranked lowly in Alexa, which should not
represent their real rankings.
Therefore, adding the domain knowledge to our
previous rankings is necessary for calculating the
authorities of financial web pages. We find some
resources about the rankings of Chinese finance
newspapers, periodicals and websites. Based on the
resources and the opinions of some domain experts,
the importance of some sources is adjusted, i.e. the
scores of some professional and important financial
websites are increased, the less important financial
websites are re-ranked lowly, and the scores of some
well-known portals are decreased, since their main
scopes are not finance. With the process of
adjustment, we are able to build a database for
source rankings. Moreover, more sources will be
added to the database with the use of our model.
Consequently, the database will contain more and
more information about sources in the domain. This
is useful knowledge for the authority calculation and
can be reused in the future. Therefore the effort of
adjustment is quite worthy. Through the adaptation
to finance domain, the importance of sources
accords more with the real situation within the
domain, with which we will acquire more accurate
result in our experiment.
4.2 Data Collection and Preprocessing
The process of data collection and preprocessing
obtains the necessary information for our authority
model, which includes link structure, source
information and related information.
The process of getting link structure includes
web page crawling, hyperlink extraction, hyperlink
filtering and link relationship establishment. The
web pages used in our experiments are crawled from
Sina Finance (http://finance.sina.com.cn/), which
contains thousands of financial news at home and
abroad. These pages form the original set for our
experiment. After the pages are downloaded from
the Internet, their contents are analyzed, and the
hyperlinks in them are extracted. In order to limit the
web pages to the finance domain and research the
relationships of financial pages, a filtering process is
done after hyperlink extraction, which restricts the
hyperlinks to Sina Finance and removes
advertisement and navigation hyperlinks. In this way
we make sure that all the web pages left are about
finance. With the hyperlink lists of the original set,
the corresponding pages of out-links are added. The
in-links that point to the original set are also taken
into consideration. These in-links are extracted from
Site Explorer of Yahoo!
(http://siteexplorer.search.yahoo.com/), and during
the extraction, the number of in-links for every web
page is limited to 50. For the new added pages of
in-links and out-links, the link relationships among
them are also established to completely form the link
structure for all the web pages.
Besides that, the source information of web
pages is also extracted from pages, and the
importance scores are obtained from Alexa and then
normalized. Then the process of adjustment is done
to source importance to add the domain knowledge.
Related hyperlinks in the web pages are also picked
up and the corresponding relationships are
established. The process of data collection and
A DOMAIN-RELATED AUTHORITY MODEL FOR WEB PAGES BASED ON SOURCE AND RELATED
INFORMATION
249

Figure 4: The process of data collection and preprocessing.
preprocessing is shown in Figure 4.
In our experiment we crawled 581 web pages
from Sina Finance at the beginning, after
establishing link relationships for these pages, the
total number of web pages raises to 22558, and there
are 860557 link relationships among the pages.
After data collection and preprocessing, our
model is applied to the web pages. The iteration of
Equation (1) converges after computing 19 times. In
the authority model the parameters are set as
follows. For Equation (1) α is 1 and β is 0.5. And in
Equation (2) a is 0.6, b is 0.3 and c is 0.1. The
analysis on choosing the values of parameters is
described in Section 4.5.
4.3 Partitioning Authorities
into Different Ranks
In practical application, users tend to accept and
favor a rank level for each web page, which is easy
and convenient for them to judge the pages.
Partitioning the authorities of web pages into
different ranks is also helpful for us to observe the
distribution of the authorities in the result, and better
evaluate the effect of our model.
We partition the experiment result into three
different ranks according to the authorities of web
pages. The first rank represents the very important
and authoritative web pages, pages which belong to
rank 2 are ordinarily authoritative, and pages in rank
3 are the least important. The partition uses a
method which is similar to k-means (Lloyd, 1982).
In the method, each authority of web pages is
assigned to the rank whose center is the nearest to it
during each iteration. The center of a rank is
represented by the average value of the authorities of
all the pages in that rank. The iteration stops when
the ranks of all web pages no longer change. This
method is able to adapt to the distribution of the
authorities of web pages and partition the web pages
into different ranks. With the method, we are able to
better analyze and evaluate our result.
4.4 Evaluation Criterion and Result
We use a manually annotated set as a standard to
evaluate the experiment result of the authority
model. We randomly select 250 web pages from our
collection, and ask three persons with finance
knowledge to score these pages independently. The
scores range from zero to 5.0. Higher scores indicate
more authoritative pages. By averaging them we get
the final score for each page. To compare our result
with the manual annotation, we need to map the
scores to the three ranks. The mapping between
scores and ranks is shown in Table 1, in which s
represents the score for a web page. The mapping is
based on the distribution of the scores in the
annotated set.
ICSOFT 2010 - 5th International Conference on Software and Data Technologies
250

Table 1: Mapping between scores and ranks.
Score Rank
2.4 <= s <= 5 1
1.2 <= s < 2.4 2
0 <= s < 1.2 3
Next, we need to compare the ranks given by our
authority model with manual annotation. We
calculate the number of pages whose ranks given by
our model are the same as the manual annotation,
and the precision of our model is calculated as
below.
 
   

(3)
Number of correct pages is the number of web pages
whose ranks given by our model are equal to the
manual annotation, and N is the cardinality of the
annotated set.
Another evaluation criterion is Average Rank
Difference (ARD), which is calculated using the
equation below.

 
∑|









|



(4)
In Equation (4), r
a
(i) represents the rank of page i
given by our model, and r
m
(i) is the rank by manual
annotation. ARD calculates the average difference of
ranks between our model and manual annotation.
With the measurement of precision and ARD, we
will have an objective evaluation of the authority
model. After calculating the scores of the pages in
the annotated set, mapping them to three ranks, and
calculating precision and ARD, the result is as
follows. Precision is 82.8%, and ARD is 0.172. The
distribution of rank difference is shown in Figure 5,
in which the horizontal axis denotes the rank
difference, and the vertical axis denotes the number
of pages. The pages whose rank differences are 0
and 1 cover the majority, which means that the result
of our model is good and acceptable.
Figure 5: The rank difference distribution for the
annotated set.
After analyzing the partitioning result and the
corresponding pages, we have the following
conclusions. The web pages which belong to rank 1
are from the important sources and have many
related hyperlinks, in rank 2 the web pages are from
less important sources and their related information
is less too. As for rank 3, the pages do not have
source information, or have less related information
and fewer link relationships in them. This result is
consistent with the design of our model, and reflects
the influences of link structure, source and related
information. These three factors are the
characteristics of web pages which are easy to be
extracted and quantitatively describe the authorities
of web pages. By considering and combining them
to our model, we are able to obtain the reasonable
result in the experiment.
Also, the pages whose positions are high in the
ranking list are often the financial news published by
important sources. These pages usually cover the
reports which most people concern. Therefore we
can find more link relationships of the pages, and
there are often more related pages. On the contrary,
web pages which are ranked lowly are some
commentary articles which express personal
opinions. The ideas of these pages are subjective,
thus their lower rankings given by our model are
quite reasonable.
4.5 Determination of Parameters
In our authority model, there are two parameters, α
and β in Equation (1), and three parameters, a, b and
c in Equation (2). The values of these parameters
will influence the effectiveness of our model. To
compare the differences of results when using
different parameters, we make several experiments
on the same collection of web pages. The
comparison of experiment results when different
parameters are set is shown in Figure 6 and Figure 7.
The experiments are made using the principle of
exhaustion, with the step length 0.1. Due to the
limits of pages, we only list a few results.
In the experiments, the best performance is achieved
when α = 1, β = 0.5, a = 0.6, b = 0.3, c = 0.1. To
achieve the best performance of our model, it is a
better way to run the authority model on a small
collection first, then applying the most suitable
parameters to the whole collection.
A DOMAIN-RELATED AUTHORITY MODEL FOR WEB PAGES BASED ON SOURCE AND RELATED
INFORMATION
251

Figure 6: Comparison of experiment result. The parameter
β changes from 0.2 to 1 when α = 1, a = 0.6, b = 0.3, c =
0.1.
Figure 7: Comparison of experiment result. Parameters a,
b and c change when α = 1, β = 0.5.
4.6 Comparing with Link Analysis
In order to illustrate the reason why we combine the
three factors together into our model, we choose
only the link structure and make the experiment on
the same collection of web pages. When only link
structure is considered in our model, it can be
viewed as the traditional link analysis algorithm.
For link analysis, the precision is 50%, and ARD
is 0.86. Comparing with traditional link analysis, our
model is able to give more reasonable authorities
and ranks. Therefore, combining source and related
information in our model is important to calculate
the authority of web pages in a specific domain.
With the authorities and ranks of the web pages,
our model provides the users with great reference
information whether to trust the pages. This is
helpful for people who read the pages, and they will
have a better judgment of the pages with our model.
5 CONCLUSIONS
In this paper we propose a domain-related authority
model which is able to calculate the authorities of
web pages in a specific domain. Three factors which
will influence the authorities of web pages are taken
into consideration, link structure, source information,
and related pages. In order to adapt to the
characteristics of the domain, we also add the
domain knowledge to the model. Experiments show
that our authority model is capable of providing
good authority scores and ranks for web pages and
facilitating people’s reading experience as reference
information. Compared with the traditional
algorithms, the authorities of our model are more
reasonable and appropriate. Therefore it has reached
our expectations and met the requirement of the task.
In the future we plan to extract some other
characteristics of web pages with domain knowledge,
and apply them to our model. We believe that with
the characteristics which are able to precisely
describe the authorities of web pages, our model will
have a better result.
ACKNOWLEDGEMENTS
This work is supported by the National Natural
Science Foundation of China (Grant No. 90818021).
We also would like to thank Naiqiao Du, Zixiao
Yang and Bowen Zhang for their participation in
manual annotation.
REFERENCES
Amento, B., Terveen, L., Hill, W., 2000. Does “authority”
mean quality? Predicting expert quality ratings of web
documents. In 23rd Annual Intl. ACM SIGIR Conf. on
Research and Development in Information Retrieval,
296-303.
Bharat, K., Mihaila, G. A., 2001. When experts agree:
Using non-affiliated experts to rank popular topics. In
Proceedings of the Tenth International Conference on
World Wide Web, 597-602.
Boldi, P., Santini, M., Vigna, S., 2005. PageRank as a
function of the damping factor. In Proceedings of the
14th International Conference on World Wide Web,
557-566.
Borodin, A., Roberts, G.O., Rosenthal, J.S., Tsaparas, P.,
2001. Finding authorities and hubs from link
structures on the World Wide Web. In Proceedings of
the 10th International Conference on World Wide
Web, 415-429.
Borodin, A., Roberts, G.O., Rosenthal, J.S., Tsaparas, P.,
2005. Link analysis ranking: algorithms, theory, and
experiments. ACM Transactions on Internet
Technology, 5(1), 231-297.
ICSOFT 2010 - 5th International Conference on Software and Data Technologies
252

Brin, S., Page, L., 1998. The anatomy of a large-scale
hypertextual web search engine. In Proceedings of the
7th International Conference on WWW.
Cohn, D., Chang, H., 2000. Learning to probabilistically
identify authoritative documents. In Proceedings of
the 17th International Conference on Machine
Learning. 167-174. Stanford University.
Cohn, D., Hofmann, T., 2000. The missing link - a
probabilistic model of document content and hypertext
connectivity. Advances in Neural Information
Processing Systems (NIPS), 13.
Eiron, N., McCurley, K.S., Tomlin, J.A., 2004. Ranking
the web frontier. In Proceedings of the 13th
International Conference on World Wide Web,
309-318.
Haveliwala, T.H., 2002. Topic-Sensitive PageRank. In
Proceedings of the 11th International World Wide
Web Conference, 517-526.
Kleinberg, J., 1999. Authoritative sources in a hyperlinked
environment. J. ACM, 46.
Lempel, R., Moran, S., 2001. SALSA: The stochastic
approach for link-structure analysis. ACM
Transactions on Information Systems, 19(2), 131-160.
Lloyd, S.P., 1982. Least squares quantization in PCM.
IEEE Transactions on Information Theory. IT-28(2),
129-137.
McSherry, F., 2005. A uniform approach to accelerated
PageRank computation. In Proceedings of the 14th
International Conference on World Wide Web,
575-582.
Nie, L., Davison, B.D., Qi, X., 2006. Topical link analysis
for web search. In Proceedings of the 29th annual
international ACM SIGIR conference on Research and
development in information retrieval, 91-98.
Yin, X., Han, J., Yu, P.S., 2008. Truth discovery with
multiple conflicting information providers on the web.
IEEE Transactions on Knowledge and Data
Engineering, 20(6), 796-808.
A DOMAIN-RELATED AUTHORITY MODEL FOR WEB PAGES BASED ON SOURCE AND RELATED
INFORMATION
253