Japan Advanced Institute of Science and Technology

Japan Advanced Institute of Science and Technology

JAIST Repository

https://dspace.jaist.ac.jp/

Title

Author(s)

Citation

Issue Date

Type

Text version

URL

Rights

Description

for Regional Information

Yousuke Ohtsuki

School of Information Science,

Japan Advanced Institute of Science and Technology

February 15, 2000

Keywords: Web Directory, Automated Editing, RegionalInformation, World Wide

Web, InformationExtraction.

As World WideWeb (WWW)becomesthe worldwideinformation medium,itis now

indispensablefor bothoÆcialorganizationsand individualstoannounce theinformation.

Inevitably, there is already a lot of useful regional information in WWW|from infor-

mation about public service for the specic regions toinformation for tourists. One can

traveleverywhere withoutwonderingwhere togoandwhat tosee ifhe can useWWW as

travel information source instead of printed travel magazine or tourist books. Regional

informationontheWWWisexpectedtobeusedaseducationtoolsatelementaryschools.

Thereare twotypesoftoolsthathelp usnd thatregionalinformation: searchengine

andwebdirectory. Bothofthem haveproblems. A searchenginereturnstoomanypages

as asearchresult. A web directoryrequires a lotof eorts tocreate and maintain.

As asolutiontothe problems, I proposea systemthat automatically generatesa web

directory for regionalinformation.

Thegenerateddirectorypresentsregionalinformationforall3427regions(prefectures,

cities,towns, andvillages)inJapan. Thedirectoryprovidestwoviews: regionalviewand

category view. In the regional view, the directory provides a page for each region. For

example, the directory provides the page that contains regional information of Ishikawa

prefecture. Inthecategoryview, thedirectorygeneratesapagethatpresentsinformation

of the specic category in one or more region according to the request. Forexample, we

can see the page about hot springs inHokuriku area.

The system consistsof the following three elements:

1. contents database

Thecontentsdatabasestoresalloftheinformationthatispresentedbythedirectory.

Copyrightc 2000byYousukeOhtsuki

Thepresentationmoduleaccepts auser'srequestandgeneratesthe pagebyretriev-

ingthe contents database.

3. automated editing module

Theautomatededitingmoduleisthe heartofthe system. Thismodulecollectsand

generates allof the informationthat is presented by the directory.

The automated editingmodule producesthe contents of the directoryfromthese two

typesof informationsources:

1. web site for aspecic region

Thistypeofsite presentsvariousinformationofaspecicregion. Atypicalexample

of this type isan oÆcialweb site of a city,town, and village.

2. web site for aspecic type of information

This type of site presents a specic typeof informationfor allregions in Japan. A

typicalexampleof this type isthe web site thatprovidespopulationof each region.

The editing process diers by the above two types of information sources. For the

rst type,the editingprocess consistsof automatic collectionof web sites and automatic

classicationofwebpages. Forthesecondtype,theeditingprocessconsistsofinformation

extractionand informationreproduction.

Because there are many web sites that present information forthe specic regions, it

is almost impossibleto collect these web sites by hand. The automatic collection of web

sites enables to automate this process. The automatic collector rst generates a set of

URLsbyarule|thenamingruleofthedomainsforcitiesandtownsinJapan. Byusinga

searchengine,thecollectorsearches thepagesthatcontainmanylinkstotheseURLs: the

pagesmay belink collectionsof regionalweb sites. The collectorextractseveryunknown

URLonthepages,and addsittotheset ofURLs. Byiteratingthisprocess, thecollector

collects many URLs of regional web sites. In the experiment, the collector collects 4012

URLs of regional sites in total; they cover 2852 regions (83.2 percent) of all regions in

Japan.

The automatic classication assigns one of eight categories to each web page in each

regionalsite. The eight categories are:

general

industry and development

festivalsand otherevents

culture,history and education

leisure and sightseeing

statistics

links

The automatic classieruses the words and phrases that are distinctive to each cat-

egory. If one of these words and phrases appears in the anchor text, the page title, or

other enhancedstrings inthe page,the classierassigns the categoryassociatedwith the

words and phrases. In the closed test, we obtained 87.9 percent recall and 87.9 percent

accuracy. In the open test, we obtained 71.4 percent recall and 76.2 percent accuracy.

From the second type of informationsource, i.e., web site for a specic type of infor-

mation, the automated editor extracts information in the form of attribute-value pairs.

Usually, these sites use the table format to present information of all regions: the table

analysisenablesextractinginformationfromthetables. Thenumericalvaluesareusedfor

informationreproduction: for example, populationdensity iscalculated from population

and area.