• Aucun résultat trouvé

Human-machine Collaboration for Enriching Semantic Wikis using Formal Concept Analysis

N/A
N/A
Protected

Academic year: 2022

Partager "Human-machine Collaboration for Enriching Semantic Wikis using Formal Concept Analysis"

Copied!
10
0
0

Texte intégral

(1)

Semanti Wikis using Formal Conept Analysis

AlexandreBlanshé,HalaSkaf-Molli,PasalMolli,andAmedeoNapoli

LORIA

Nany,Frane

{firstname.lastname}loria.fr

Abstrat. Semanti wikis are new generation of ollaborative tools.

They allow to embedsemanti annotationsinthe wiki ontent. These

annotationsallowtobetterorganizeandstruturethewikiontents.It

isthenpossibleforuserstobuildknowledgeunderstandablebyhumans

andomputers.Bythisway,mahinesareallowedtoprodueorupdate

semantiwikipagesashumansando.Inthispaper,weproposeanew

smartagent based onFormalConeptAnalysis. Thissmartagent an

omputeautomatiallyategorytreesbasedondenedsemantiproper-

ties.Inordertoreduehuman-mahineollaborationproblems,humans

justvalidatehangesproposedbythesmartagent.Adistributedversion

ofwikiisusedtoensureonsistenyoftheontentduringthevalidation

proess.

Keywords. FormalConeptAnalysis,SemantiWiki,Human-Mahine

Collaboration

1 Introdution

Semanti wikis are new generation of ollaborativetools [1,2,3,4℄. They allow

toembedsemantiannotationsin thewikiontent.Theseannotationsallowto

betterorganizeandstruturethewikiontents.Semantiwikisallowmassol-

laborationforreatingandemergingontologialresoures.Theyguidetheusers

frominformalknowledgeontainedindoumentstomoreformalstrutures.

Semantiwikisallowuserstobuildknowledgeunderstandablebyhumansand

omputers.Bythisway,theyalsoallowmahinestoprodueorupdatesemanti

wikipagesashumansando.Thisopenstheopportunitytoonsidermahinesas

newmemberofommunitiestoprodueandmaintainknowledge.Consequently,

suhsmartagents anreduesigniantlytheoverheadofommunitiesinthe

proessofontinuouslyknowledgebuildingandorrethumans errors.

In[5℄,authorsoupledaase-basedreasonerwithasemantiwiki.Thease-

based reasoneran enrih thewikis with newsemantipages and thus an be

onsideredasasmartagent.Aspointedoutin[5℄,human-mahineollaboration

an lead to unstable system if not managed. For example, if humans hange

theategorytreeused bythe ase-basedreasoner,thease-basedreasoneran

(2)

sis(FCA)[6℄.Thissmartagentanomputeautomatiallyategorytreesbased

ondened semantiproperties.Bythisway,theFCAsmartagentleverageshu-

mans from these tasks. Inorder to redue human-mahine ollaborationprob-

lems, humans just validate hangesproposed bytheFCAsmartagent.This is

ahievedusing theDSMW[7℄semantimediawikiextension.

Thepaperisorganizedasfollows.Setion2introduestheFCAframework.

Setion 3 shows how the FCA smart agent is used to enrih thewiki. Setion

4 details the validation proess. The last setion onludes and points future

works.

2 Formal Conept Analysis

Inthispaper,wepresentasmartagentthatenrihawikibasedonalassiation

method. Atually,anylassiationmethodsmightbeused. WehooseFormal

Conept Analysis (FCA) beause it extrats oneptsorganizedinto alattie,

whih isinteresting forthenavigationinto thewiki. In thissetion, webriey

introdueFCA.

Formal Conept Analysis [6℄ is a lassiation method allowingto build a

onept lattie where onepts are omposed of an intent, a maximal set of

attributes,andanextent,amaximalset ofobjetssharingtheattributes.

Aontext

K

reliesonasetofobjets

G

,asetofattributes

M

andarelation

betweenobjetsofattributes

I ⊆ G × M

. Consideringanobjet

g ∈ G

andan

attribute

m ∈ M

,

(g, m) ∈ I

meansthat

g

hastheattribute

m

.

Aontext anbevisualizedasabinary table.Table1showsa(simple)ex-

ampleofontextaboutanimals.Thereareveattributesthatdesribeanimals.

Animalsmayhavehair,feather,wings.Theymightbreathin airorwater.Ob-

jetsareanimals:bat,bird,atandsh.Inthetable,arossinoneellindiate

theanimalhastheorrespondingattribute.

Hashair Hasfeather Haswings Breatheinair Breatheinwater

Bat Ö ÖÖ

Bird ÖÖÖ

Cat Ö Ö

Fish Ö

Table1.Exampleofontext(animals)

(3)

FCA allows to build onepts organized into a lattie. A onept

C 1 = (A 1 , B 1 )

is dened byan extent

A 1

(a set of objets) and anintent

B 2

(a set

of attributes that dene the onept).If

C 2 = (A 2 , B 2 )

is a suboneptof

C 1

(denotedby

C 2 ⊑ C 1

),then

A 2 ⊆ A 1

and

B 1 ⊆ B 2

.Thetoponept

ontains

all the objetsand usually its intent is empty (unless an attribute is present

in eahobjet). Thebottom onept

is dened byallattributes but usually

ontainsnoobjets(unlessanobjethasallattributes).

On gure1 is shown the onept lattie of the ontext of table 1. On the

graph,everynodeisaonept.Alinkbetweentwonodesindiatesasubsumption

relation(aoneptisasuboneptofanotheronept).Theintentofaonept

iswrittenonagraybakground,theextentonwhitebakground.

Fig.1.Galois lattiebasedontheontextfromtable1

3 Wiki Enrihment

3.1 Priniples

Wedevelopedamethodthatreorganizestheategoriesofthewikiaordingto

theresultofFCA.Anewwikiwillbereatedwiththesamepagesandproperties,

butdierentategories,basedonthelattieofonepts.

Thenewategorieswillbereatedbasedonthepreviousones,andonseman-

tilinks betweenpages.Usefulategorieshumanusersdidnotreate mightbe

disovered.Itisevenpossibletostartawikiwithoutreatinganyategoriesbut

onlysemantilinks betweenpages,andthenletthesmartagentbuildtheate-

gories,basedonthesemantilinks.Thenewategoriesfailitatethenavigation

(4)

Eahategorymapsone(andonlyone)onept:themostgeneralonepton-

taining the ategory in its intent (the attribute onept). Eah onept maps

zero,oneorseveralategories.Ifaoneptmapsasingleategorytheategory

willbepreserved.Ifaoneptmapstwoategoriesormore,itmeanstheseat-

egoriesare identialandshould bemerged(howeverthis aseisveryunlikely).

Ifaoneptdoesnotmapanyategory,anewategorywillbereated.

Currently,theenrihmentisperformedbyaJavaappliationthataessthe

ontentofthewikiandreateanenrihedversionofit.

3.2 Case study

The method presented in this paper will be illustrated by a wiki onerning

aademis.Herewepresenttheinitialontentofthewiki.Wehavethefollowing

(user-dened)ategories:

Category:Professor;

Category:Topi;

Category:Course;

Category:Level whih ontains two subategories: Category:Master 1

LevelandCategory:Master 2 Level.

Wealsodened twoproperties:

Property:isTaughtBy,thedomain isaourse,therangeaprofessor;

Property:isAbout,thedomainisaourse,therangeatopi.

Finally,weaddedpagesinthewiki:

Prof. SmithandProf. JonesintheProfessorategory;

Artifiial Intelligene, Software Engineering and Networks in the

Topiategory;

Knowledge Disovery, in the Course and Master 1 Level ategories,

thispage hastwosemantilinks isAbout:Artifiial Intelligeneand

isTaughtBy:Prof. Smith;

Semanti Wiki, in the Course and Master 2 Level ategories, this

page has two semanti links isAbout:Artifiial Intelligene and

isTaughtBy:Prof. Smith;

Semanti Web, in the Course, Master 1 Level and Master 2 Level

ategories, this page has two semanti links isAbout:Artifiial

IntelligeneandisTaughtBy:Prof. Smith;

Design Patterns, in the Course and Master 1 Level ategories, this

page has two semanti links isAbout:Software Engineering and

isTaughtBy:Prof. Jones;

Network Administration,in the Courseand Master 1 Levelategories,

thispagehastwosemantilinksisAbout:NetworksandisTaughtBy:Prof.

Jones;

IPv6 Protool, in the Courseand Master 2 Level ategories,this page

(5)

FCA anbeapplied ontheontentofthe wiki.Objetstobelassiedbythe

FCAalgorithmarethestandardpagesofthewiki.

Thedesriptionofapageisomposedoftwoparts:theategoriesitbelongs

toandthesemantipropertiesithas(inourrstprototype,weonlyonsidered

wikipropertiesoftypePage).Eahofthesetwopartsallowtobuildaontext.

Weanombinethesetwoontextbyapposition.

Based on the ontent of the wiki, as desribed above, we an reate the

ontextshownontable2.Whenappliedtothisontext,FCAreturnsthelattie

shownongure2.

Table 2.Contextbasedonthewiki

Professor Topi Course Level Master1Level Master2Level isTaughtBy:Prof.Smith isTaughtBy:Prof.Jones isAbout:ArtiialIntelligene isAbout:SoftwareEngineering isAbout:Networks

Prof.Smith Ö

Prof.Jones Ö

ArtiialIntelligene Ö

Networks Ö

SoftwareEngineering Ö

KnowledgeDisovery ÖÖÖ Ö Ö

SemantiWeb ÖÖ Ö Ö Ö

SemantiWiki ÖÖÖÖ Ö Ö

DesignPatterns ÖÖÖ Ö Ö

IPv6Protool ÖÖ Ö Ö Ö

NetworkAdministration ÖÖÖ Ö Ö

Intheasestudy, asoneanseeongure2,fouroneptsmath oneate-

gory: Professor,Topi,Master 1 Level,andMaster 2 Level.Oneonept

mathestwoategories:CourseandLevel.Alltheotheroneptsdonotmath

anyategoryatall.

How to reate the new ategories depends on the number of ategories

mathed by eah onept. Depending on that number dierent methods are

used. However,no ategories are reatedfor the twoonepts

and

, as

alwaysontainsallpagesand

doesnotontainanypage.

(6)

3.4 Preserving of an originalategory

Ifaoneptmathesoneandonlyoneategory,thisategorywillsimplybepre-

servedintheenrihedwiki.ThisistheaseoftheategoryTopi,forinstane.

Atually,inmostases,alltheoriginalategoriesarepreserved.

3.5 Category merging

If aonept mathes two ategories or more, a new ategory is reated. This

newategorywillmergetheontentoftheoriginalmathingategories:textof

eahpagesareonatenatedtogether.Adefaulttitle isgiventotheategory.

Categorymergingshouldberare.It onlyhappensiftwoormoreategories

alwaysappearin theexatsamepages.Thiswouldhappenif severalusersuse

dierenttermsforthesameonept.Bitbybit,afteranumberofwikiedition,

these dierent ategories will appear in all the same pages and then will be

mergedbytheFCA.

Thisis thease ofthetwoategories Courseand Level.Having these two

ategories is due to a naming problem. The enrihed wiki has now only one

ategoryforthisonept.

3.6 New ategories

(7)

apagebelongsto twoategoriesormore;

severalpageshavingsomeidentialproperties.

A ategoryaboutourses on software engineering has been reated,based

onthesemantirelationin thepageDesign Patterns.Also,aategoryabout

ourses available for both Master 1 and Master 2 students has been reated,

Semanti Webisapageofthisategory.

3.7 Category enrihment

Whateverthereationmethodofaategory,allthenewategoriesareenrihed

withnewtextontent,basedonproperties.SenteneslikeThepagesbelonging

tothisategoryseemstohaverelation

T

withthepage

P

.wouldbeappendedin

thepage.Thiswillhelphumanuserstounderstandthemeaningoftheategory.

For instane, the ategoryof ourses about software engineering will on-

tainthe sentene Thepagesbelongingto this ategoryseemsto haverelation

Property:isAboutwiththepageSoftware Engineering.,asadesriptionof

theategory.

4 Validation

4.1 Validationby humanusers

Aftertheenrihment,newategoriesneedtobevalidatedbyhumanusers.Some

merged ategories might be spit, some new ategories removed. Also, human

usersshould editall theategories:default titles should behanged into more

relevantones,textshould berened. Wewill presentthreeexamplesofvalida-

tion.

TherstoneonernsthetwoategoriesCourseandLevelthathavebeen

merged.Havingthistwoategorieswasamistake.Humanuserswillaknowledge

that and rename the merged ategory Course. They will also rename two of

thesubategoriesMaster 1 CourseandMaster 2 Courseto makethemmore

intelligible.

Another example onerns a new ategory that has been reated based

on the semanti relation in the page Design Patterns with a default name

(Category:New Category 42,forinstane).Asexplainedinpreviously,thenew

ategorywillontainatextdesribingsomepropertiesoftheonept.Ahuman

user will understand that this ategoryontains ourses about software engi-

neering and will rename it onsequently. Thesame thing will be done for the

ategoryaboutoursestaughtbyProf.Jones.

The last example onerns a subategory of Master 1 Courseand Prof.

Jones' Course.One mightonsider this ategoryto be irrelevant,or at least

not useful. A humanuserwould deide to removethis ategoryfrom the wiki

(8)

Fig.3.Man-mahineollaborationproess

Inorderto ensureonsistenyofthe data,weusedadistributed wiki. Two

semantimediawiki sites are synhronizedwith theDSMW extension 1

[7℄ (see

gure3).

TherstoneistheSemantiWiki1wiki.Humansaessthiswikiasusual.

FromthisSemantiWiki1,theFCAsmartagentreatesthelattiein the

SemantiWiki2 site.

Humanuserswillthenhektheontentofthisseondwikisite,orretand

renetheontent.

Next,theyanpushtheontentofSemantiWiki2onapushfeed.

Finally, administrator of SemantiWiki1 anpull validated modiations

fromSemantiWiki2intoSemantiWiki1.

This senario demonstrates how the DSMW extension anbe used to im-

plement proesses. In this ase, a simple proess allows validation of hanges

produed by the FCA smart agent and avoids the problem of instability of

human-mahineollaboration.

4.3 Enrihed wiki ontent

After validation, here is the ontent of the enrihed wiki (SemantiWiki1 in

gure3)intheasestudy:

Category:Professor,ontainspagesaboutProf.SmithandProf.Jones;

1

(9)

SoftwareEngineering;

Category:Course;

Category:Master 1 Course,asubategoryof Category:Course;

Category:Master 2 Course,asubategoryof Category:Course;

Category:Artifiial Intelligene Course, a subategory of

Category:Course, the page indiates that Prof. Smith is teahing all

theoursesinthis ategory;

Category:Prof. Jones' Course,asubategoryof Category:Course;

Category:Master 1 Artifiial Intelligene Course, a subategory

of Category:Master 1 Courseand Category:Artifiial Intelligene

Course,ontainsthepageaboutKnowledgeDisovery;

Category:Master 2 Artifiial Intelligene Course, a subategory

of Category:Master 2 Courseand Category:Artifiial Intelligene

Course,ontainsthepageaboutSemantiWiki;

Category:Master 1 and 2 Artifiial Intelligene Course, a sub-

ategory of Category:Master 1 Artifiial Intelligene Course and

Category:Master 2 Artifiial Intelligene Course, ontains the

pageaboutSemantiWeb;

Category:Networks Course, a subategory of Category:Prof. Jones'

Course;

Category:Software Engineering Course, a subategory of

Category:Prof. Jones' Course and Category:Master 1 Course,

ontainsthepageaboutDesignPatterns;

Category:Master 1 Networks Course,asubategoryofCategory:Master

1 Courseand Category:Networks Course,ontainsthe page aboutNet-

workAdministration;

Category:Master 2 Networks Course,asubategoryofCategory:Master

2 Courseand Category:Networks Course,ontains the pageaboutIPv6

Protool.

5 Conlusion and future work

Semantiwikis allowusersto build knowledge understandableby humans and

omputers.Bythisway,theyalsoallowmahinestoprodueorupdatesemanti

wikipagesashumansando.Thisopenstheopportunitytoonsidermahinesas

newmemberofommunitiestoprodueandmaintainknowledge.Consequently,

suhsmartagents anreduesigniantlytheoverheadofommunitiesinthe

proessofontinuouslyknowledgebuildingand orrethumans errors.

In this paper, we proposed a new smart agent based on Formal Conept

Analysis. This smart agent allows to reorganize the wiki: new ategories are

omputedand pagesare plaedintothese newategories.This allowsabetter

organizationoftheontentandfailitatethenavigationin thewiki.

Therefatoringproessneedsto bevalidated byhumanusers.Consisteny

(10)

validation.

Thispaperpresentedanearlywork,andmoreresearhhavetobedoneinthe

future.Clearly,ifappliedonarealwiki,amethodsuhasFCAwouldproduea

largeamountofonepts,anditwouldbyimpossibleforhumanuserstovalidate

anyone of them. Some ltering methods should be used to preventirrelevant

ategoriestobeadded,basedonthenumberofinstanesin aategoryorother

riteria.

UsingRelationalConeptAnalysisinsteadofFCAshouldprovideinteresting

results.Otherlusteringmethods willalsobeonsidered.

Intheurrentversionofourmethod, humanusershaveafeedbakfromthe

smartagent,theywilltakeintoonsiderationthenewategoriesthathavebeen

reated. However, the smart agent does not have afeedbak from the human

users: ifa ategoryhasbeen rejetedduring thevalidation proess, thesmart

agent will reate it again when the proess will be reiterated. To avoid this

problem, thesmartagenthastobehistory-aware andusetheinformation of

themodiationbyhumanusersduringthevalidationproess.

6 Aknowledgments

This researh waspartof the CyWiki projet, funded bythe Université Henri

PoinaréofNany.

Referenes

1. Völkel,M.,Krötzsh,M.,Vrandei,D.,Haller,H.,Studer,R.:Semantiwikipedia.

In: WWW '06: Proeedings of the 15th international onferene on World Wide

Web.(2006)585594

2. Shaert,S.:IkeWiki:Asemantiwikiforollaborativeknowledgemanagement.In:

1stInternationalWorkshoponSemantiTehnologiesinCollaborativeAppliations

(STICA06),Manhester,UK.(2006)

3. Bua, M., Ereteo, G., Faron-Zuker, C., Gandon, F., Sander, P.: SweetWiki:A

semantiwiki.JournalofWebSemantis,speialissueonWeb2.0andtheSemanti

Web6(1)(2008)

4. Krötzsh,M.,Vrandei,D.,Völkel,M.,Haller,H.,Studer,R.:Semantiwikipedia.

Journal ofWebSemanti5(4)(2007)251261

5. Cordier, A., Lieber, J.,Molli, P.,Nauer, E., Skaf-Molli, H., Toussaint, Y.: Wiki-

taaable:Asemantiwikiasablakboardforatextualase-basedreasoningsystem.

In: 4th Workshop on Semanti Wikis (SemWiki2009), held in the 6th European

SemantiWebConferene.(2009)1832

6. Ganter,B.,Wille,R.:FormalConeptAnalysis,MathematialFoundation.Springer

(1999)

7. Rahhal, C., Skaf-Molli, H., Molli, P., Weiss, S.: Multi-synhronous ollaborative

semanti wikis. In:10th International Conferene on Web Information Systems-

Wise2009.Volume5802ofLetureNotesinComputerSiene.(2009)

Références

Documents relatifs

Using equivalence relations de- fined for each single column, we are able to translate the bicluster mining problem in terms of the partition pattern structure framework.. We show

We have developed a method using formal concept analysis and semantic query expansion that assists the user in defining their seed query and in refining the expanded search space

The current work semantic web resources ranking is by adapting and modifying pagerank algorithm used in classical search engines.. ReConRank [11], TripleRank[12] adapt

This approach is capable of finding strict dependencies between lines of source code covered in successful and failed runs.. The output is a logical expression which allows to debug

The CLR method consists of three main steps: i) construction of the formal context of documents-terms and building of the corresponding concept lattice ii) insertion in the lattice of

The talk can be seen as a roadmap for how FCA can be used in other social network analysis domains such as social media or the twitter sphere highlighting both the strengths

After introducing the basic definitions needed from category theory and formal concept analysis, in this paper we have studied two different product construc- tions in the

For example, let us consider the SPARQL query given in Listing 1.1, for all the entities of type Automobile manufac- tured by Lamborghini, annotated as “Sport_cars” and