REAL-TIME DISCOVERY OF CURRENTLY AND HEAVILY

VIEWED WEB PAGES

Kazutaka Maruyama

Information Processing Center, The University of Electro-Communications

Chofugaoka 1-5-1, Chofu, Tokyo, 182-8585, Japan

Kiyotaka Takasuka, Yuta Yagihara, Satoshi Machida, Yuichiro Shirai, Minoru Terada

Dept. of Information and Communication Engineering, The University of Electro-Communications

Chofugaoka 1-5-1, Chofu, Tokyo, 182-8585, Japan

Keywords:

Peer-to-peer, clustering, web browsing, browser extension.

Abstract:

The amount of information on web increases explosively, so it is difﬁcult for web users to ﬁnd web pages they

want. There are some approaches to resolve this problem, such as semantic web which make web information

systematic, the improvement of search engines’ algorithm, and so on. Dealing with web as a huge database,

these technologies works well, however, they cannot provide any useful solutions to get hot news which

expands quickly to the world because of their time lag.

In this paper, we propose a system whose users can know currently and heavily viewed web pages. The

key features of this system are as follows: (1) to ﬁnd hot news in web, (2) to provide recommendations to

users without any content analysis, and (3) to apply the system to other communication tools like IM as their

infrastructure to ﬁnd appropriate contact targets. We describe our policy of the system implementation and

show the result of a pilot experiment with a pilot implementation.

1 INTRODUCTION

In the recent web, the amount of information on web

increases explosively, for example consumer gener-

ated media like blog, and the pace seems to be ac-

celerated much more in future. Users know that

web pages they want to ﬁnd exist somewhere in web

world, but they can ﬁnd only part of the target pages

and always worry about the lack of useful ones.

Two major approaches to this problem are seman-

tic web and the improvement of search engines. Web

contents with rich meta data and search engine algo-

rithms such as PageRank can produce more appropri-

ate answers to users’ queries. However these solu-

tions have three problems. First, it is considerably

difﬁcult to add appropriate and sufﬁcient meta data to

all existing web contents, and also probably all ones

generated in future. Second, the intervals of crawl-

ing by search engine robots and of updating the in-

dexes do not become much shorter and are not in real-

time. Third, the solution cannot create any clusters of

the users, not the contents. We call them communi-

ties. In real life, communities exist everywhere such

as in schools, ofﬁces, web bulletin boards, and so on.

We can always talk with each other about various hot

topics in such communities. For instance, a person

who knows music scene thoroughly may tell us about

not only recent trends of CD charts, but also a rumor

about unpublished next iPod. Search engines provide

results only to explicitly given terms by users, but we

cannot ﬁnd other topics which implicitly relate to the

terms because they have no user communities.

We propose an infrastructure to ﬁnd currently and

heavily viewed web pages in real-time by exchanging

viewed URLs at the client side. Three advantages of

this architecture are as follows.

1. We can quickly discover web pages which are at-

tracting considerable attention, such as a new beta

service of Google disclosed tonight.

2. We can know web pages which most of friendly

users view. The term of friendly users means ones

who usually view the same web pages as we have

viewed.

3. Communication tool developers or researchers can

use our system as an infrastructure of new tools

they are developing, such as IM, because it cre-

ates communities of users who usually view the

same web pages, and probably have the same pref-

erences.

Since our system does not need any meta data, all

the existing web contents in the world are available

352

Maruyama K., Takasuk K., Yagihara Y., Machida S., Shirai Y. and Terada M. (2006).

REAL-TIME DISCOVERY OF CURRENTLY AND HEAVILY VIEWED WEB PAGES.

In Proceedings of WEBIST 2006 - Second International Conference on Web Information Systems and Technologies - Internet Technology / Web

Interface and Applications, pages 352-359

DOI: 10.5220/0001250503520359

 SciTePress

and the content providers and consumers can use the

system without any special actions (the ﬁrst problem

above is solved). The system provides currently and

heavily viewed web pages mostly in real-time, be-

cause exchanging URLs need not crawl or index any

web pages (the second is solved). In addition, it can

very easily introduce the mechanism correspondent

to mouth-to-mouth advertising in real life by creat-

ing user communities with only usual actions of web

users, that is to view web pages in their browsers (the

third is solved).

The remainder of this paper is organized as follows.

Section 2 shows the features of our proposal with an

use case scenario. In section 3, we discuss the losses

against gains of the implementation strategies of the

client and server side subsystems, and describe the

implementation of our currently developing system.

Section 4 shows the result of a pilot experiment with

a pilot system. Other applications of our system are

described in section 5 and the related works are dis-

cussed in section 6. Section 7 and 8 describe conclu-

sion and future works respectively.

2 FEATURES OF OUR SYSTEM

In this section, we show the features of our proposal in

the view point of users and the current user interface

through an use case scenario.

2.1 An Use Case Scenario

Alice is a very common user of web, every day crawls

her favorite news sites and blogs, and looks for a

reputation of a restaurant which she found on her

way home by using search engines. She changed

her jobs recently and was not satisﬁed because she

could not ﬁnd interesting topics in web well, which

a well-informed colleague in the former ofﬁce told

her, through news sites, blogs or search engines. She

could not ﬁnd news sites or blogs just suitable for her

interests and was tired of reading so many RSS feeds.

Search engines are very powerful tools when she is

looking for topics with explicit purpose, but they do

not help her to look for recently popular topics or

somehow nice events.

One day, she began to use a system mentioned in a

certain blog. It was said that users of the system ex-

change URLs they are just viewing. The information

exchanged is only URL, so she did not have to take

care of copyright violation in contrast to ﬁle sharing.

Following the instruction of the ofﬁcial site of the sys-

tem, she installed a tiny extension for Mozilla Firefox,

an add-on program for the web browser, to her own

Firefox. She clicked an icon added by the extension

at the lower right of Firefox, then a pop-up menu ap-

peared there. It controls the appearance of the sidebar

described below, and enables or disables the mecha-

nism of sending out URLs she is viewing. Turning

on the sidebar, she saw the appearance of her Firefox

such as ﬁgure 1. The sidebar lying on the left side of

the window has a few buttons to list the correspondent

rankings, and clicking these buttons causes switches

of the list displayed in the pane. Entries of blogs or di-

aries in social network services viewed heavily were

listed at the top of the lists. Alice attempted to partic-

ipate in the project of exchanging URLs and enabled

the mechanism of sending out URLs by the menu of

the lower right icon. Therefore URLs she is viewing

became to be shared in the world.

One week after, URLs listed in the sidebar be-

came suitable for Alice’s preference. The project web

page said that the system connects users to each other

with fewer hops, who tend to view the same web

pages, and consequently provides more appropriate

rankings. She viewed web pages in the same way

as before, but she became to know some cool prod-

ucts which she had never known or some recipes of

healthy diets. Using the new way of retrieving in-

formation different from news sites, blogs or search

engines, she became to use web more efﬁciently and

had fun with various hot topics.

2.2 Summary of Features

We described a basic use case of our proposal through

Alice’s experience. The summary of the features is as

follows.

• A quite new channel for information retrieval is es-

tablished. In real life, for instance, a crowd on

a street shows that there is something interesting

such as a street performer. While the same method

of information retrieval did not exist so far, it is in-

troduced into the world of web.

• The proposed system provides real-time informa-

tion. Users need not to wait for a blog entry to be

submitted by the owner who saw a news source.

For example, we can reach an entry of Slashdot, to

which comments are increasing.

• Communities are created. The server side system

connects friendly users to each other with fewer

hops by using URLs they have viewed. In other

words, unfriendly users are kept away from each

other and S/N ratio of the provided rankings would

be improved.

• All users have to do is an installation of a single

extension once, in contrast to social bookmarkings

which need their users to continue submitting each

entry for exchanging URLs. The users of our sys-

tem can provide useful information of web pages

only by viewing web pages in their usual way and

can get useful URLs from others in the world.

REAL-TIME DISCOVERY OF CURRENTLY AND HEAVILY VIEWED WEB PAGES

353

Figure 1: Alice’s Firefox with our extension.

The use case described in this section includes the

functions which are not implemented yet, but shows

the user experience through our proposal. Another

aspect of the system, an infrastructure of communica-

tion tools, is described in section 5.

3 SYSTEM STRUCTURE

In this section, we describe the implementation strate-

gies of the client side and the server side subsystems,

discuss the losses against gains of the strategies, and

describe the current system implementation.

3.1 Extension vs. Proxy

One of the important behavior is to capture URLs an

user is viewing in real-time without any interferences.

There are two implementation choices: (1) add-on

program for web browsers such as extension or tool-

bar, and (2) web proxy server.

The best way for users is web proxy. All users have

to do is to set up their browsers so that the proxy is

used, then all the web behavior of the users is cap-

tured by the proxy in real-time. The proxy would be

placed at school or ofﬁce, or be installed as a quite

small Java program without cache facility into each

user’s PC. The greatest beneﬁt of this choice is that

the installation procedure of users is the minimum.

When a proxy is used by many users, for example for

all students in a college, the single client side subsys-

tem can capture so many users’ web behavior. How-

ever, this solution has a disadvantage of collecting too

many meaningless URLs. When an user clicks a cer-

tain link, his browser, and the proxy of course, must

retrieve many resources in the clicked document, such

as images, style sheets, ﬂash applications including

advertisements. The ranking of captured URLs will

place the Google logo image at the top of the list. Of

course such the result is not expected. Using Ref-

erer ﬁelds in HTTP or analyzing transferred HTML,

clicked images may be distinguished from ones in-

cluded clicked web pages. But the analysis makes the

system complex and the performance down, conse-

quently, the users must become irritated.

Another implementation choice is extension or

toolbar, which have the opposite features. These add-

on programs work as a part of a web browser and can

capture explicitly clicked URLs only. This feature is

an obvious advantage against the proxy. On the other

hand, the extension has a few disadvantages. First,

users have to install the extension or toolbar, even if it

is a very small program and only two or three clicks

ﬁnish the installation. Second, at least two implemen-

tations for Firefox and Internet Explorer are required

in order to support over 90% of web browsers in the

world. Since a Firefox extension works on all plat-

forms supported by Firefox, such as Windows, Linux,

WEBIST 2006 - WEB INTERFACES AND APPLICATIONS

354

Figure 2: Three buttons of the sidebar.

Figure 3: Pop-up menu.

Mac OS X and so on, a toolbar for Windows IE and

the extension can cover almost all the client environ-

ment.

As described in section 2, we chose the exten-

sion/toolbar implementation, but the toolbar for IE is

not implemented yet. The appearance of Firefox with

the extension is shown in ﬁgure 1. In the current ver-

sion, the sidebar on the left of the window has three

buttons (ﬁgure 2), and three rankings appear by the

correspondent buttons, i.e. recent top 1000, weekly

chart and monthly chart. An additional ranking of

URLs to which accesses increase suddenly would be

added. Figure 3 shows the pop-up menu from the icon

at the lower right of the window. Sending URLs is

disabled in the default setting.

3.2 Server Centric vs. Peer-to-Peer

The extension is the client side half of our system.

The server side one collects URLs sent from the client

side. There are also two choices: (1) server centric

and (2) peer-to-peer.

Obviously, the server centric approach is simpler.

Powerful servers with high speed CPUs and large

disks in a data center could deal with URLs sent from

all users in the world. For example, Google and Ya-

hoo! start their services which save the search histo-

ries of each user. They must use such many servers

and disks. In this approach, it is easier to exchange

users’ web behavior and to analyze relevance among

users. On the other hand, the disadvantage is that the

large system of many servers is too expensive and re-

quires a large amount cost for maintenance.

Another approach is P2P. The users install a tiny

program running on background written in Java into

their PC and the program works as a node of the

Server

CGI

programs

Apache

web

server

Clients

Viewed

URLs

Firefox

web

browser

Firefox

extension

POST URLs

GET Rankings

Figure 4: Current system structure.

P2P system. The client side programs, extensions

or proxy servers, send URLs to and receive rankings

from the node. The large system need not be con-

structed. However, a simple and ﬂat P2P network

such as Gnutella cannot deal with so many URLs

from so many users in the world and cannot connect

friendly users to each other. Clustering P2P nodes

and a layered architecture, such as Skype(Skype - The

whole world can talk for free, nd) or Winny(Kaneko,

2005), could resolve this problem.

At present, we choose the server centric approach

because of the ease of development. The server con-

sists of some CGI programs, the extension sends

URLs to it as POST data of HTTP and receives the

rankings from it by GET method of HTTP. Note that

there is an explicit interface between the client side

subsystem and the server side one. The interface is

deﬁned as URLs and CGI parameters over HTTP.

When we replace the server with P2P based subsys-

tem, if the interface between the extension and the

server is preserved, no changes are required for the

extension. In addition, the interface of HTTP is ex-

tremely suitable for web browser extensions.

3.3 Current System Structure

The current system structure is shown in ﬁgure 4. The

client side subsystem consists of Firefox web browser

and the extension and the server side subsystem con-

sists of Apache web server and some CGI programs

written in Perl. The extension sends clicked URLs

to the server by POST method and the server stores

them into disks. When the user clicks a button of

the sidebar shown in ﬁgure 2, the extension sends a

GET requests to the CGI program correspondent to

the clicked button, then receives the list of the rank-

ing.

The system produces any explicit recommenda-

tions of web pages, because the ranking of the re-

cent top 1000 shows the recommendations from users

of the system. The weekly and monthly charts show

the recommendations from them in the week and the

month respectively. The ranking of URLs to which

accesses increase suddenly, which is still developing,

REAL-TIME DISCOVERY OF CURRENTLY AND HEAVILY VIEWED WEB PAGES

355

Internet Explorer with our

toolbar (developing)

Pilot system client

Firefox with our extension

Client side subsystem Server side subsystem

P2P

network

Server with CGI programs

Figure 5: Three kinds of clients and two of servers.

shows the recommendation at the moment.

Prior to developing the extension, we developed

a pilot implementation of the client side subsystem,

the top in ﬁgure 5 (Takasuka et al., 2005). In fact,

the server side subsystem was developed for the pi-

lot client. The current client, the middle in ﬁgure 5,

preserves the interface between the pilot client and

the server, and replaces the pilot one without any

changes of the CGI programs. The coming toolbar

of IE shown in ﬁgure 5 would work together these ex-

isting clients.

Clustering users are not introduced yet into our sys-

tem. First, users of our system are still around ten.

Because the extension has some problems, our sys-

tem is not released yet to the outside of our laboratory.

Second, the clustering is relevant to the identiﬁcation

of URLs. This point is described in section 8.

4 PILOT EXPERIMENT

In order to examine the usefulness of our proposal,

we implemented a pilot system to do a pilot exper-

iment. In the pilot system, a client side subsys-

tem is a simple and tiny web browser written in C#

with IE component on Windows (shown in ﬁgure 5),

and a server side subsystem consists of the CGI pro-

grams described above. The simple browser has for-

ward/backward buttons, a simple bookmarklet and a

facility of sending URLs to the server side subsystem.

In this experiment, nine student users used the pi-

lot system for four days. The total accesses from the

browser to web pages were 448 times. 70 accesses in

them were from the rankings of the simple browser.

Table 1 shows the result of the questionnaire. Each

item is rated in one to ﬁve, one means the best and

ﬁve the worst. The usefulness of the system we pro-

pose is evaluated comparatively highly. In addition,

note that the rate of the psychological barrier in shar-

ing web history is neutral. While the experiment is

really small, the proposed system was accepted favor-

ably, and the increase of the number of users would

make the beneﬁt more and the barrier lower.

5 OTHER APPLICATIONS OF

OUR SYSTEM

While we do not implement the facility to connect

friendly users yet, once the facility is introduced, it

could be used as an infrastructure of communication

tools. Two application examples are described below.

The ﬁrst is a chat among the users who are viewing

the same web page. In contrast to Instant Messenger,

of which the users designate their partners to talk to

in advance, the chat we propose connects users who

are viewing or have been viewing the same web page

and helps them to talk to each other about the web

page or some topics about it. In web bulletin boards

such as Slashdot, the users who have never met before

can talk to each other, but the place for the discussion

must exist already and they cannot know whether the

user who wrote a certain comment is viewing the page

now. Our proposal provides the interchange between

the web users through ordinary web pages. The users

need not look for a thread, in which they are inter-

ested, and can talk to friendly users only by viewing

their favorite web pages usually. Improving the chat

system, as IM users can know that their friends have

logged in, the users could know the others are view-

ing a certain web page. The improvement makes web

pages rendezvouz.

The second is a P2P based bulletin board system.

It is difﬁcult for P2P based BBSs to gather friendly

users in the same thread because of the structure of

P2P. However, using our system as its infrastructure,

the P2P based BBS could gather many users, who

tend to view the same web pages and may have the

similar preferences.

The core application is the rankings extension de-

scribed in section 3, which sends data to web his-

tory exchange layer in order to connect friendly users

to each other and receives the rankings from the

layer (ﬁgure 6). The other two applications described

in this section only receive location information of

friendly users from the layer.

WEBIST 2006 - WEB INTERFACES AND APPLICATIONS

356

Table 1: Evaluation of the pilot system.

Questions (1:good, 5:bad) Av.

Serviceability of the browser 3.1

Usefulness of sharing web history

2.0

To watch URLs others view is helpful

1.7

There is no psychological barrier in sharing web history

2.8

Web history exchange layer

(Server centric or P2P)

Rankings

(extension)

Chat

Bulletin

board

system

TCP/IP and lower layers

Figure 6: Web history exchange network as a basis of com-

munication tools.

Other projects of communication tools would use

the layer as their infrastructure. Whenever a quite

new network service starts, there are two problems,

the scalability of the server and the gathering of the

sufﬁcient initial users. Our system is useful for new

network services or communication tools. Singh et

al.(Singh et al., 2001) said that P2P networks can pro-

vide a substrate for community-based service loca-

tion. The system we propose just can be applicable

to it.

6 RELATED WORKS

Webmemex(Alaniz et al., 2003) provides the web

page recommendations based on sharing web behav-

iors among the well-known users. They register each

other in their contact list of Yahoo! Messenger, prob-

ably also well-known in real life. The proxy server

gathers the viewed URLs and the back end system

computes the relevance among the web pages by

HTML analysis. When an user clicks a certain link,

the system recommends some web pages which are

related to the clicked web page and are viewed by

other users of the group. It is similar to our proposal,

but there are some essential differences.

• Our system deals with only URL strings, not the

contents of the web page. The vocabulary in the

web world is changing very quickly, so it is too

difﬁcult for such the analysis based system to con-

tinue following the changes, especially in some

languages which do not place a space between the

words such as Japanese. Since the data format used

in the web world is also changing quickly, it seems

not to be realistic to analyze the various contents of

the web pages in the various areas.

• We aim for the discovery of the web pages which

are attracting considerable attention with little time

lag. The computation of the relevance of the con-

tents among the web pages causes the opposite

result to the purpose and it is suitable for post-

processing. It should be dealt with as a different

challenge.

• Our system does not limit the range of exchanging

URLs. Limiting the range to the well-known users

leads to the consequence of losing chance of com-

ing across information in different areas. In addi-

tion, since the well-known users strongly connect

to each other both in real and virtual life, their psy-

chological barriers avoid exchanging URLs.

Social bookmarking services such as

del.icio.us(del.icio.us, nd) aim for the similar

goal. When an user ﬁnds others who bookmarks

some web pages in which the person are also

interested, the person implicitly gets their recommen-

dations of hot topics or good web pages through their

bookmarks. The ﬁrst disadvantage is that the social

bookmarking users have to explicitly go into action

to add the URLs they prefer. The second is that the

users looking for hot topics or good web pages have

to look for friendly users too.

Fast el al.(Fast et al., 2005) introduce a social net-

working in order to cluster the nodes of P2P ﬁle shar-

ing based on the downloaded music ﬁle categories of

the P2P users. In comparison between P2P ﬁle shar-

ing and P2P URL sharing, search queries in ﬁle shar-

ing correspond to advertising of URLs in URL shar-

ing, and downloaded ﬁles correspond to really viewed

URLs by the users who receive the others’ advertise-

ments. The number and frequency of the transmis-

sion between the queries and the advertisements seem

to be considerably different, and URL sharing mostly

deals with statistical information, for instance many

users are viewing a certain URL, in contrast to ﬁle

sharing which ﬁnds the real entities, music ﬁles. The

different knowledge from the paper about P2P clus-

tering would be found.

REAL-TIME DISCOVERY OF CURRENTLY AND HEAVILY VIEWED WEB PAGES

357

Winny(Kaneko, 2005), a P2P ﬁle sharing software

with over 200,000 users in Japan, uses three key-

words for a background ﬁle downloading process.

The keywords are also used for P2P clustering, but

the users have to set the keywords in advance. In

the Winny network, the ﬁles opened to be shared by

a certain node is cached at the other nodes, such as

Freenet(Clarke et al., 2002), and the network load is

heavy. Therefore, in order to realize more efﬁcient

ﬁle sharing, Winny constructs the layered P2P net-

work based on each node’s connection speed, for ex-

ample, up to 64kbps, xDSL or ﬁber. The P2P URL

sharing distributes much smaller data such as URLs,

so it have to make much of whether the reboot inter-

vals of the nodes are long or whether the nodes have

a ﬁxed global IP address, rather than the connection

speed.

7 CONCLUSION

Exchanging URLs which users are viewing in real-

time brings hot topics, which are attracting consider-

able attention, without any keywords explicitly desig-

nated. The effective way to realize it is that the client

side subsystem should be implemented as an add-on

program for web browsers, and the server side as P2P.

In the current version of our system, the server side

is implemented as the server centric CGI programs,

but the P2P could take the place of it by preserving

the interface, which consists of HTTP and CGI, be-

tween the server side and the client side. The pilot

experiment showed the usefulness of our proposal. In

addition, the system for exchanging URLs could be

applied to other communication tools as their infras-

tructure.

8 FUTURE WORKS

The top pages of portal sites tend to be placed at the

higher rank. This problem could be resolved through

the user interface of the extension. The users disable

the appearance of such entries in the rankings, for ex-

ample by a right click on a target entry, and exchange

the disabled portal sites via the server side subsystem.

The strategy of clustering users is quite important

and difﬁcult. Creating clusters is an effective way

to reduce the number of connections, the trafﬁc on

P2P network, and the computation of the similarity of

users. The separation of clusters, however, may make

users lose opportunity to be offered interesting web

pages(Linden et al., 2003). In ordinal recommenda-

tion systems such as e-commerce marketing systems,

each user’s purchases in a day may be usually less

than ten items, even if they are heavy users of the sys-

tem. But web pages each user views in a day are at

least ten pages, and heavy web users view more than

100 pages. In addition, there are web users obviously

more than e-commerce site users. Reduction by clus-

ters must be introduced for effective exchange of web

histories. Parameters for establishing clusters, such

as the number of connections of each node, the limit

of hops of web history transfer, and so on, could be

found through simulation by using huge amount of

proxy logs of our university.

The identity of URLs is also important. While

CMSs generate different URLs to each blog entry,

known as permalink, visitors can read the entries not

only in their permalinks but also in the summary

pages of recent entries, of the day, of the month and

so on. Consequently, a certain entry has many URLs

where it can be read. It is desirable that these URLs

are dealt with as the same. This problem could be re-

solved by the introduction of scores between similar

URLs. The URLs which indicate the same entry are

similar, because most CMSs add the date of the entry

to its permalink. In addition, the similarities between

URLs would help us to cluster users. Users who often

see different entries of the same blog should be placed

in the same cluster.

In the view of the privacy, it is important not to

send URLs which should not be shared. Our exten-

sion does not send any URLs in the default setting and

explicitly shows its setting, whether sending URLs is

turned on or off, by the correspondent icons. On the

other hands, the web pages without appropriate access

controls, such as password authentications or Limit

directives of web servers, must be seen by outsiders.

It is equivalent to being opened even if the URLs are

not opened. This problem is known as “Google hack-

ing” among security experts. Appropriate settings of

access controls would make the problem trivial.

Spammers may attack this system by sending

URLs which they want to advertise. This problem

is known as “shilling attacks”. However, since the

information for clustering is URLs themselves, the

spammers may be classiﬁed to a spammers’ cluster

and cannot inﬂuence the ordinal users. The clustering

of spammers would be inspected through simulation.

REFERENCES

Alaniz, A., Truong, K. N., and Antonio, J. (2003). Automat-

ically Sharing Web Experiences through a Hyperdoc-

ument Recommender System. In Proceedings of the

14th ACM Conference on Hypertext and Hypermedia,

pages 48–56.

Clarke, I., Miller, S. G., Hong, T. W., Sandberg, O., and

Wiley, B. (2002). Protecting free expression online

with Freenet. IEEE Internet Computing, 6(1):40–49.

WEBIST 2006 - WEB INTERFACES AND APPLICATIONS

358

del.icio.us (n.d.).

http://del.icio.us/.

Fast, A., Jensen, D., and Levine, B. N. (2005). Creating So-

cial Networks to Improve Peer-to-Peer Networking. In

Proceedings of the 11th ACM SIGKDD International

Conference on Knowledge Discovery in Data Mining,

pages 568–573.

Kaneko, I. (2005). Technology of Winny. Ascii Publishing.

(In Japanese).

Linden, G., Smith, B., and York, J. (2003). Amazon.com

Recommendations: Item-to-Item Collaborative Filter-

ing. IEEE Internet Computing, 7(1):76–80.

Singh, M. P., Yu, B., and Venkatraman, M. (2001).

Community-based Service Location. Communica-

tions of the ACM, 44(4):49–54.

Skype - The whole world can talk for free (n.d.).

http://www.skype.com/.

Takasuka, K., Shirai, Y., Maruyama, K., and Terada, M.

(2005). A web browser that shares browsing histories.

In FIT2005, pages 355–356. (In Japanese).

REAL-TIME DISCOVERY OF CURRENTLY AND HEAVILY VIEWED WEB PAGES

359