Anotherpossibilityistoentersomealreadyknownimplicationsbeforestartingtheexploration.Theseimplications,theuseralreadyknowstobevalid,



Similar documents
V. Adamchik 1. Graph Theory. Victor Adamchik. Fall of 2005

Path Querying on Graph Databases

MATHEMATICS Unit Decision 1

Decision Mathematics D1 Advanced/Advanced Subsidiary. Tuesday 5 June 2007 Afternoon Time: 1 hour 30 minutes

Answer: (a) Since we cannot repeat men on the committee, and the order we select them in does not matter, ( )

Social Media Mining. Graph Essentials

Outline. NP-completeness. When is a problem easy? When is a problem hard? Today. Euler Circuits

Graph Theory Problems and Solutions

Network/Graph Theory. What is a Network? What is network theory? Graph-based representations. Friendship Network. What makes a problem graph-like?

On the Number of Planar Orientations with Prescribed Degrees

CS311H. Prof: Peter Stone. Department of Computer Science The University of Texas at Austin

1.1. The Goal of Clustering

Handout #Ch7 San Skulrattanakulchai Gustavus Adolphus College Dec 6, Chapter 7: Digraphs

On the k-path cover problem for cacti

Linear Coding of non-linear Hierarchies. Revitalization of an Ancient Classification Method

MASTER OF SCIENCE EDUCATION

3. Eulerian and Hamiltonian Graphs

Quantum Monte Carlo and the negative sign problem

Introduction to Graph Theory

Lecture Notes on GRAPH THEORY Tero Harju

Mathematics for Algorithm and System Analysis

Graph Theory: Penn State Math 485 Lecture Notes. Christopher Griffin

Graph Theory Lecture 3: Sum of Degrees Formulas, Planar Graphs, and Euler s Theorem Spring 2014 Morgan Schreffler Office: POT 902

WOLLONGONG COLLEGE AUSTRALIA. Diploma in Information Technology

Discrete Math for Computer Science Students

measured (empirical) data from CCC. The modelled values are of value for all (ecosystem specific modelling work deposition)

A Study of Sufficient Conditions for Hamiltonian Cycles

Math 179: Graph Theory

Network Metrics, Planar Graphs, and Software Tools. Based on materials by Lala Adamic, UMichigan

IE 680 Special Topics in Production Systems: Networks, Routing and Logistics*

A 2-factor in which each cycle has long length in claw-free graphs

DEPARTMENT OF INFORMATION TECHNOLOGY SEMESTER: IV

Mathematical Problem Solving for Elementary School Teachers. Dennis E. White

Discrete Mathematics Problems

Transportation Polytopes: a Twenty year Update

13 Reading Device/PLC from Database

8.1 Min Degree Spanning Tree

Graph Classification and Easy Reliability Polynomials

A Fast Algorithm For Finding Hamilton Cycles

Notes on NP Completeness

This means that any user from the testing domain can now logon to Cognos 8 (and therefore Controller 8 etc.).

Krishna Institute of Engineering & Technology, Ghaziabad Department of Computer Application MCA-213 : DATA STRUCTURES USING C

New Approach of Computing Data Cubes in Data Warehousing

Exercises of Discrete Mathematics

COPYRIGHTED MATERIAL. Contents. List of Figures. Acknowledgments

Computer Algorithms. NP-Complete Problems. CISC 4080 Yanjun Li

Open Problems in Quantum Information Processing. John Watrous Department of Computer Science University of Calgary

Lavastorm Analytic Library Predictive and Statistical Analytics Node Pack FAQs

Factor Models for Gender Prediction Based on E-commerce Data

Seavus Group All rights reserved

CGMgraph/CGMlib: Implementing and Testing CGM Graph Algorithms on PC Clusters

John Engbers Curriculum Vitae

The Butterfly, Cube-Connected-Cycles and Benes Networks

Lecture 30: NP-Hard Problems [Fa 14]

CIS 700: algorithms for Big Data

CSC 373: Algorithm Design and Analysis Lecture 16

Total colorings of planar graphs with small maximum degree

CROPS: Intelligent sensing and manipulation for sustainable production and harvesting of high valued crops, clever robots for crops.

Getting Started With Delegated Administration

HOLES 5.1. INTRODUCTION

SUPPORTED ACTIVE DIRECTORY TOPOLOGIES BY LYNC 2013

UNIFIED BIJECTIONS FOR MAPS WITH PRESCRIBED DEGREES AND GIRTH

The Role of Mathematics in Information Security Education

A Turán Type Problem Concerning the Powers of the Degrees of a Graph

SHARP BOUNDS FOR THE SUM OF THE SQUARES OF THE DEGREES OF A GRAPH

How To Create A Graph From A Graph With A Powerpoint Powerpoint Toolbox

arxiv: v1 [cs.cc] 17 Jul 2015

Discrete Mathematics & Mathematical Reasoning Chapter 10: Graphs

Topological Properties

UPPER BOUNDS ON THE L(2, 1)-LABELING NUMBER OF GRAPHS WITH MAXIMUM DEGREE

Montefiore Portal Quick Reference Guide

FileZilla: Uploading/Downloading Files to SBI FTP

CWSRF Project Descriptions and Examples for Green Project Reserve

Boulder Dash is NP hard

Findings from the 9 th Annual MetLife S tudy of Employee Benefits Trends A Blueprint for the New Benefits Economy

Using SAS ACCESS to retrieve and store data in relational database management systems

DeparTmenT of mathematics 200 Albert W. Brown Building (585) ; Fax: (585) Major in Mathematics (46 credits)

Topology. Shapefile versus Coverage Views

The Institute and Puerto Rico: A Beautiful Relationship

Creating a File Geodatabase

Transcription:

BackgroundImplicationsandExceptions AttributeExplorationwith jectsinaspeciedcontext.thisknowledgerepresentationisespeciallyuseful Summary:Implicationsbetweenattributescanrepresentknowledgeaboutob- Schlogartenstr.7,D{64289Darmstadt,stumme@mathematik.th-darmstadt.de TechnischeHochschuleDarmstadt,FachbereichMathematik cspringer-verlagberlin{heidelberg1995 erdstumme offormalconceptanalysisthatsupportstheacquisitionofthisknowledge.fora plicationsbetweenattributesofthiscontexttogetherwithalistofobjectswhich whenitisnotpossibletolistallspeciedobjects.attributeexplorationisatool speciedcontextthisinteractiveproceduredeterminesaminimallistofvalidim- backgroundimplications)andallvalidimplications.thelistofimplicationscan setofimplicationsthatllsthegapbetweenpreviouslygivenimplications(called besimpliedfurtherifexceptionsareallowedfortheimplications. arecounterexamplesforallimplicationsnotvalidinthecontext.thispaperdescribeshowtheexplorationcanbemodiedsuchthatitdeterminesaminimal 1.Introduction speciedcontext.thisknowledgerepresentationisespeciallyusefulwhenit isnotpossibletolistallspeciedobjects.attributeexploration(cf.anter Implicationsbetweenattributescanrepresentknowledgeaboutobjectsina (1987),Wille(1989))isatoolofformalconceptanalysisthatsupportsthe acquisitionofthisknowledge.formalconceptanalysiswasintroducedin Wille(1982)andhasgrownduringthelastfteenyearstoanusefultoolin determinesaminimallistofimplicationsthatissucienttodeduceallvalid dataanalysis. Foraspeciedcontexttheinteractiveprocedureofattributeexploration implicationsbetweentheattributesofthecontexttogetherwithalistof hasalreadysomeideahowtheattributesarerelated.beforestartingthe objectasacounterexample. Usually,theuserdoesnotstartsuchanexplorationfromscratch,buthe objectswhicharecounterexamplesfortheimplicationsthatarenotvalidin usercaneitheracceptasuggestedimplicationorhemustsupplyanew thecontext.asaninteractiveprocedure,theprogramsuggestsimplications thenumberofremainingpossibleimplications. explorationhecanenteralistofobjectswhichmaysignicantlydecrease totheuserwhichdonotcontradicttoalreadygivencounterexamples.the Anotherpossibilityistoentersomealreadyknownimplicationsbeforestartingtheexploration.Theseimplications,theuseralreadyknowstobevalid,

InP.Burmeistersimplementation(1987),itisalsopossibletoenterbackgroundimplications.Thisprogramhoweverdeterminesaminimalsetof areherecalledbackgroundimplications.thispaperdescribesageneral- ofimplicationswhich togetherwiththebackgroundimplications is izationofattributeexploration,thatitisabletodetermineaminimallist sucienttodeduceallvalidimplications. thisprogramdoesnotdependinanywayonthebackgroundimplications onestartswith.intheapproachpresentedinthispaperthebackground usedtodecreasethenumberofquestionstotheuser.theresultinglistin implications regardlessofthebackgroundimplications.theyareonly implicationsareusedtominimizeonlythenumberofadditionallyneeded describeacontextbyaminimalnumberofimplications.theremaybemore Inthenextsectionthebasicdenitionsofformalconceptanalysisarerecalledandexplainedbyanexample,beforeimplicationsincontextsand implicationsisdescribed.itisilluminatedbyanexampleinthelastsection, sectiontheinteractiveprocedureofattributeexplorationwithbackground thenotionsofcompletenessandirredundancyareintroduced.inthethird implicationsifsomeofthemareobvious. implications.thisisbasedonthebeliefthatitisnotalwaysthebestto theresultoftheexploration. 2.ImplicationsofContextsandtheL-Duquenne- wherewealsodiscuss,howtheadmissionofexceptionscanfurthersimplify Firstwebrieyrecallthebasicdenitionsofformalconceptanalysis(cf. Denition:Wecall(;M;I)acontext,whereandMaresetsandIis anterandwille(1995))andgiveanexample. arelationbetweenandm(i.e.im).theelementsofandm read:\theobjectghastheattributem". arecalledobjectsandattributes,respectively,andgim(:()(g;m)2i)is uigues-basis ForeverysetAofobjectswedenethesetA0:=fm2MjgIm forallg2agofallattributessharedbyallobjectsina.duallytheset B0:=fg2jgImforallm2Bgisthesetofallobjectshavingall Nowaconceptofthecontext(;M;I)isapair(A;B)withA,BM, attributesinbm. Inmumandsupremumintheconceptlatticearecalculatedasfollows: A0=B,andB0=A.ThesetAiscalledtheextentoftheconcept,theset Btheintent.Thehierarchicalsubconcept-superconcept-relationisgivenby (A1;B1)(A2;B2):()A1A2(()B1B2).Thesetofallconcepts whichiscalledtheconceptlatticeof(;m;i)andisdenotedbyb(;m;i). ofacontext(;m;i)togetherwiththisorderrelationisacompletelattice t2t(at;bt)=(\ ^ t2tat;([ t2tbt)00);_ t2t(at;bt)=(([ t2tat)00;\ t2tbt):

malcontextshowninfig.1.itsobjectsarethegraphs1upto18and relationshipscontainedintheunderlyingdatacontext:weconsiderthefor- itsattributesaretenattributesofundirectedgraphs(cf.wilson(1975)): Thefollowingexampleshowshowtheconceptlatticeunfoldstheconceptual connected,disconnected,bipartite,complete,completebipartite,tree,forest, planar,eulerian,hamiltonian.thiscontext(togetherwithalistofimplications)istheresultoftheattributeexplorationthatisdescribedinthe (g),whichisdenedastheconceptwiththesmallestextentcontaining g,islabeledwith\g".dually,foreveryattributemitsattributeconcept extentwhicharelinkedtoitbyadescendingpathanditcontainsallthose (m),whichisdenedastheconceptwiththesmallestintentcontaining bedeterminedinthediagram:aconceptcontainsallthoseobjectsinits m,islabeledwith\m".thentheextentandintentofeveryconceptcan itsextentandtheattributesconnected,complete,eulerian,andhamiltonian attributesinitsintentthatarelinkedtoitbyanascendingpath.the nected),(planar)and(bipartite),andtheotherby(connected),(planar) rightmostconceptinfig.2forexamplehasthegraphs3,14and18in Inthediagramonecanseetwocubesatthetop:oneisspannedby(discon- initsintent. and(bipartite).thisindicatesthatinbothcasesthethreeinvolvedat- Thedominatingpartinthelatticeliesbetween(connected)and(13).It tributesareindependent. isthedirectproductofa6-element\ladder"witha4-element\rectangle", butitcanalsobeseenas4-dimensionalhypercubesthataregluedtogether ateightvertices.theupperoneliesbetween(connected)and(15)and thirdsection. Itissucienttolabelthelinediagramnotwiththecompleteconcepts,but onlywiththeattributesandobjects:foreveryobjectg,itsobjectconcept and18. rian),and(hamiltonian).thisshowsthat,forconnectedgraphs,thefour attributesplanar,bipartite,eulerian,andhamiltonianareindependent. pointismissing itismarkedinthediagramwithalittledotleftof(15). Inthelowerhypercube(between(bipartite)^(connected))and(13)one ThisindicatesthateverycompletebipartiteHamiltonianplanargraphisalso Eulerian.Infactthereexist(uptoisomorphism)onlytwosuchgraphs,13 isspannedby(planar)^(connected),(bipartite)^(connected),(eule- Denition:AnimplicationbetweenattributesinMisapair(X;Y)of implicationx!yisvalidinacontextkifitisrespectedbyeveryobject intent.theimplicationisthencalledanimplicationofthecontextk.an subsetsxandyofm.itisdenotedbyx!yandisread\ximplies MthatrespectsLalsorespectsX!Y. TrespectsasetLofimplicationsifitrespectseveryimplicationinL.An implicationx!yisentailedbyasetlofimplicationsifeverysubsetof Y".AsubsetTofMrespectstheimplicationifX6TorYT.Theset

disconnected bipartite completebipartite tree connected planar Eulerian forest 1 Hamiltonian 10 2 11 3 12 4 1 2 13 5 14 6 15 7 16 8 17 9 18 3 4 19 20 5 6 7 8 9 Lemma1AnimplicationX!YisvalidinacontextKifandonlyif 10 Figure1:Contextofgraphs 11 12 13 14 B(K):Inourexamplefcomplete,Euleriang!fHamiltoniangisanimplicationofthecontext,whichcorrespondstotheequality(complete)plicationscanequivalentlybeunderstoodasequalitiesintheV-semilattice betweentheattributesofthecontext,becausethesetofallintentsisexactly thelargestclosuresystemonmthatrespectsalltheseimplications.theim- Thestructureofaconceptlatticeisalreadydescribedbyallimplications YX00.ThenitisalsorespectedbyeveryconceptintentofK. 15 16 17 18 19 lattice. (Eulerian)=(complete)^(Eulerian)^(Hamiltonian)intheconcept 20

14 15 11 8 10 7 18 2 20 19 5 9 complete 1 13 4 12 3 6 connected 17 16 disconnected bipartite planar tree forest Hamiltonian Eulerian complete bipartite Figure2:ConceptlatticeofthecontextinFig.1 Someoftheseimplicationsmaybeknowninadvance,andinthispaper theywillbereferredtoasbackgroundimplications.inaformalsenseevery implicationofacontextcanbeabackgroundimplication.thequestionis nowhowtodescribethestructureoftheconceptlatticewithimplications inthemostecientwaywhenbackgroundimplicationsaregiven.weare lookingforaminimallistthatis\llingthegap"betweenthebackground implicationsandallvalidimplications. Denition:LetKbeanitecontextandLasetof(background)implicationsofK.AsetBofimplicationsofKiscalledL-complete,ifevery implicationofkisentailedbyl[b.itiscalledl-irredundantifnoimplicationa!b2bisentailedby(bnfa!bg)[l.al-basisisa L-completeandL-irredundantsetofimplicationsofK.IfLisemptythen Biscalledcomplete,irredundant,andabasis,respectively.AsubsetPof Miscalledpseudo-intentofKifP6=P00andifforeverypseudo-intentQ withqptheinclusionq00pholds.

J.-L.uiguesandV.Duquenne(1986)showthatB:=fP!P00jPisa backgroundimplicationsisdenotedbyp7!p:=p[pl[pll[:::with Duquenne-uigues-basis.WeobtainaL-basisbygeneralizingthisdenition. pseudointentgisabasisof(;m;i)ifmisnite.thisbasisiscalled XL:=X[SfBMjAX;A!B2Lg. willbenite. Denition:TheclosureoperatoronthesetMofattributesinducedbythe foreveryl-pseudo-intentqwithqp,theinclusionq00pholds. AsubsetPofMiscalledL-pseudo-intentofKifP=P6=P00andif, Withoutfurthermentioning,allsetsofattributesconsideredinthefollowing ThesetBL:=fP!P00jPisaL-pseudo-intentgofimplicationsiscalled Proof.Obviously,allimplicationsinBLareimplicationsofK.Weprove L-Duquenne-uigues-basis. Theorem2BLisaL-basisofK. LetP!P002BL.WeshowthatP!P00isnotentailedby(BLn fp!p00g)[lbecauseprespectsallimplicationsin(blnfp!p00g)[l ThisisacontradictionbecauseTdoesnotrespectthisimplication. T6=T00.ThenTisaL-pseudo-intentbydenitionandsoT!T002BL. FurthermoreTrespectsQ!Q00foreveryL-pseudo-intentQT.Suppose thatblisl-completebyshowingthateverysubsettmrespectingall (andprovesothatblisl-irredundant):asp=p,itclearlyrespectsall implicationsinl[blisanintent:ast!tisentailedbylwehavet=t. Duquenne-uigues-basisisB=fcd!abcd;b!ab;ad!abcd;ac!abcdg.For A!A.ThecontextinFig.3showsthatingeneralthisisnotthecase.The pseudo-intentswithandthendeletingthe(trivial)implicationsoftheform OnemayaskifitispossibletogettheL-pseudo-intentsbyjustclosingthe implicationsinl.forq!q002blnfp!p00gwithqpwehave Q00P,asPisaL-pseudo-intent.HencePrespectsalsoQ!Q00. thebackgroundimplicationl:=fcd!agwegetthel-duquenne-uiguesbasisbl=fb!ab;ad!abcd;ac!abcdgwhilefp!p00jpispseudo-intent acquisitiontoolthatcanbeusedtodeterminetheduquenne-uigues-basis B.anter(1987)presentsattributeexplorationasaninteractiveknowledge 3.AttributeExplorationwithBackgroundImplications isnotl-irredundant. withp6=p00gadditionallycontainsacd!abcd.ingeneraltheresultingset ofacontextthatiseithertoolargeforacompleteinputintothecomputer orthatisevennotcompletelyknown.itisbasedonhisnext-closure- Algorithmthatecientlycalculatesclosuresystems.

1 2 3abcd TheAttributeExplorationprocedurecanbemodiedsuchthatitcanbe 4 setlofbackgroundimplications.thereforeweproceedsimilartoanter closuresystemonthesetmofallattributes: usedtodetermineinteractivelythel-duquenne-uigues-basisforagiven (1987).FirstweshowthatthesetofallintentsandL-pseudo-intentsisa Figure3: c Lemma3Let(;M;I)beacontext,letLbeasetofimplicationsof andq6p.thenp\qisanintent. (;M;I),andletPandQbeintentsorL-pseudo-intentswithP6Q Proof.PasQandthereforealsoP\QrespectallimplicationsinL[BL exceptp!p00andq!q00.becauseofp6p\qandq6p\qthe ofsimplicityweassumethatm:=f1;:::;ng. NextweintroducethelecticalorderonthesetofsubsetsofM.Forthesake setp\qrespectstheseimplications,too.henceitmustbeanintent. Corollary4ThesetofallintentsandL-pseudo-intentsofanitecontext A<B:()(9i2BnA:A\f1;:::;i?1g=B\f1;:::;i?1g)for Denition:ThelecticalorderonP(M)isdenedby (;M;I)isaclosuresystemonM;withtheclosureoperatorX7!X:= A;BM. X[X[X[:::,whereX:=X[SfBMjA!B2BL;AXg. ForA;BMandi2Mwedene A<iB:()(i2BnAandA\f1;:::;i?1g=B\f1;:::;i?1g) subsetbmthelecticallynextintentorl-pseudo-intentisthesetbi, andai:=((a\f1;:::;i?1g)\fig). lastintentorl-pseudo-intentism. whereiisthemaximalelementinmnbwithb<ibi.thelectically Theorem5ThelecticallyrstintentorL-pseudo-intentis;.Foragiven TheNext-Closure-algorithmofB.anter(1987)listsallclosedsetsofa closuresystemonanitesetinthelecticalorder.inthenexttheoremitis appliedtotheclosuresystemofallintentsandalll-pseudo-intents: a 1 b 2 d 3 4

ThistheoremprovidesthecentralpartoftheAttributeExplorationwith backgroundimplications,whichwedescribenow:wewanttodetermine thel-duquenne-uigues-basisofacontext(;m;i)(whichisapriorinot Hofobjects.ThesetHmayalsobeempty. completelygiven)forasetlofbackgroundimplicationsofthecontext. Thealgorithmstartswithapartialcontext(H;M;I\(HM)forasubset Algorithm.Setk:=1andBL:=;. (1)DeterminethekthL-pseudo-intentPkof(H;M;I\(HM)byapplyingTheorem5.IfMisreached(asintent)thenSTOP.BListhen thel-duquenne-uigues-basis. (2)Asktheuser:\IstheimplicationPk!P00 (3)otostep(1). {Iftheansweris\No",thenaskforanobjectgthatdoesnot {Iftheansweris\Yes",thenaddPk!P00 respectthisimplication.therowfgg0thenalsohastobeentered bytheuser.addgtoh. kvalid?" ktoblandincreasek. lecticallyrstl-pseudo-intentsof(h;m;j).let(;m;i)beanitecontext Theorem6Let(H;M;J)beanitecontextandletP1,:::,Pkbethek changewhenanobjectisaddedthatrespectsallpreviouslyacceptedimplicationsandallbackgroundimplications: ThealgorithmiscorrectbecausethelecticallyrstL-pseudo-intentsdonot L-pseudo-intent. PJJ L-pseudo-intentsof(;M;I). fggirespectsallpi!pjj withhandj=i\(hm),inwhichalltheimplicationsinl[fpi! Proof.Fori=1;:::;kwehavePII islecticallylessthanpj,theassertionisaconsequenceofthedenitionof iji=1;:::;kgarevalid.thenp1,:::,pkarealsotheklecticallyrst i.aseveryl-pseudo-intentqpjof(h;m;j) i=pjj ibecauseforeveryg2theset setm:=fconnected(conn),disconnected(disc),bipartite(bip),complete InthissectionweseehowthecontextinFig.1isproduced.Westartwiththe Eulerian(eul),Hamiltonian(ham)gofattributesandwanttoknowwhich (comp),completebipartite(cbip),tree(tree),forest(for),planar(plan), 4.AnExplorationofraphs Astheclassofundirectedgraphscontainsinnitelymanyisomorphism implicationsbetweentheseattributesarevalidforallundirectedgraphs. classes,thereisnopossibilitytodeterminetheduquenne-uigues-basisfor

Furthermoreweknowthatatreeisjustdenedasaconnectedforestand thatacompletebipartitegraphisalwaysbipartite.thisjustiesthefollowingbackgroundimplications: Whenwelookatthelistofattributesthenweseethatconnectedanddisconnectedarecontradictingeachother(i.e.,nographcanhavebothattributes). graphs.thegraphstheusergivesascounterexamplesduringtheattribute explorationarejustthesetypicalgraphs. theinnitecontext(;m;i)directly.onehastoworkwithsome\typical" anemptyseth.inthefollowingtheattributesappearinginthepremiseof Wedonothaveanyobjectsatthebeginning,sotheexplorationstartswith animplicationwillnotbelistedintheconclusionagain. fconn,discg!m fconn,forg!ftreeg fcbipg!fbipg ftreeg!fconn,forg TherstL-pseudo-intentistheemptyset.Thereforethedialoguestarts NowHcontainstheobject1.InthisenlargedcontexttherstL-pseudointentisstilltheemptyset,butonthisstepwehave;00=fconn,plan, withthequestion: hamg.q:is;!fconn,plan,hamgvalid? Q:Is;!Mvalid? A:No.1hastheattributesconn,plan,ham. 3isaddedtoH.InthecontextwithH=f1;2;3gtheemptysetis 2isaddedtothesetH. Q:Is;!fplangvalid? A:No.2hastheattributesdisc,bip,for,plan. anintent.thenextl-pseudo-intentisfhamg. A:No.3hastheattributesconn,comp,eul,ham. Theimplicationfhamg!fconngisaddedtoBLwhichwasemptyuptonow. Q:Isfhamg!fconngvalid? A:Yes. Q:Isfeulg!fconn,comp,hamgvalid? A:No.4hastheattributesconn,bip,cbip,plan,eul. :::

andthefollowingimplicationsareaccepted: Duringtheexplorationthegraphs1to20aregivenascounterexamples fconn,bip,tree,for,plan,hamg!fcomp,cbip,eulg fconn,bip,tree,for,plan,eulg!fcomp,cbip,hamg fconn,comp,eulg!fhamg fdisc,bip,cbipg!ffor,plang fcompg!fconng fhamg!fconng fconn,bip,cbip,plan,hamg!feulg feulg!fconng fconn,bip,compg!fcbip,tree,for,plang fforg!fbip,plang ThesetenimplicationsconstitutetheL-Duquenne-uigues-basisBL.Every implicationthatisvalidinthecontextcanbededucedfromthemandthe fourbackgroundimplications.theduquenne-uigues-basisconsistsofall andthefourthbackgroundimplicationandthetwoimplicationsftreeg! implicationsinthel-duquenne-uigues-basisandadditionallyoftherst fconn,bip,for,plangandfconn,bip,for,plang!ftreeg.inthisexample thecardinalityofthel-duquenne-uigues-basisisjustthedierenceofthe asexceptionsthatcontradictanimplicationandare(uptoisomorphism) andkeepinmindtheexceptions.forexamplewecanregardallgraphs cases,wemayconrmsomeimplicationsthataretrueforalmostallgraphs Duringtheexplorationtherearesomeimplicationsthatcanbedeniedby cardinalitiesoftheduquenne-uigues-basis,butingeneralitmaybelarger. Thebeginningoftheexplorationdialogueremainsunchanged.Therstdifferenceappearswiththequestion:\Isfcompg!fconn,eul,hamgvalid?", because6is(uptoisomorphism)theonlyconnectedcompleteplanarhamiltoniangraphwhichisnoteulerianandisthereforenotallowedascounterexample.howevertheimplicationhastobedenied:raph06infig.4 servesasnewcounterexample.thenextsuggestionfcompg!fconn,hamg fconn,eul,hamg!fcompg,because14isanexceptioninthesensedened of14thegraph014willbeusedasacounterexamplefortheimplication willbeacceptedwiththeexception7,whichistheonlycompletegraph thatisnothamiltonian.inthiswaytheexplorationcontinues.instead above. ThisapproachyieldsthefollowinglistofimplicationsthatisaL-basisfor allgraphsexceptfor6,7,13,14,and18.behindeveryimplication uniqueinhavingexactlytheirattributes. thegeneralstructureofgraphtheorywithoutbotheringwithpathological onlyonecounterexample(uptoisomorphism).ifwewanttodetermine arelisteditsexceptions.

disconnected bipartite completebipartite tree connected planar Eulerian forest 014 06 Figure4:Additionalgraphsfortheexplorationwithexceptions fdisc,bip,cbipg!ffor,plang fcompg!fconn,eul,hamg fhamg!fconng feulg!fconng fforg!fbip,plang (6;7) 14 Hamiltonian completeplanarhamiltoniangraphs:14and18.that18istheonly fconn,bip,tree,for,plan,hamg!m fconn,bip,tree,for,plan,eulg!m fconn,comp,plan,eul,hamg!m (14;18) HamiltoniantreeandtheonlyEuleriantreeisexpressedby7thresp.8th The6thimplicationindicatesthatthereexist(uptoisomorphism)onlytwo fconn,bip,cbip,plan,hamg!feulg fconn,bip,comp,eul,hamg!m (13;18) implication.itisalsotheonlyhamiltonianbipartitecompletegraph(10th implication). (18) TheresultingconceptlatticeisshowninFig.5.Theimplicationsvalidin 7,13,14,and18.Theconceptlatticeoftheseexceptionsisshown thislatticeareexactlythosewhicharevalidforallgraphsexceptfor6, infig.6.inparticularonecanseeinthediagramthatallexceptionsare connectedplanargraphs. atedbysomeconceptsisinteractivelydetermined. AttributeexplorationdeterminestheV-semilatticethatisgeneratedbythe attributeconcepts.inpremiseandconclusiononlyconjunctionsofattributes areallowed.disjunctionsbecomeinvolvedindistributiveconceptexploration (cf.stumme(1995)),wherethecompletedistributivelatticethatisgener- 6

disconnected bipartite connected planar 11 9 complete bipartite Eulerian Hamiltonian 10 complete 6 12 8 3 16 14 5 15 2 19 Figure5:Conceptlatticeresultingoftheexplorationwithexceptions 1 forest 4 17 tree 20 planar Figure6:Conceptlatticeoftheexceptions connected complete bipartite Hamiltonian complete bipartite forest Eulerian tree 6 7 14 13 18 disconnected

Kontexte.TechnischeHochschuleDarmstadt(Latestversion1995forAtariST References: Burmeister,P.(1987):ProgrammzurformalenBegrisanalyseeinwertiger anter,b.(1987):algorithmenzurbegrisanalyse.in:b.anter,r.wille,k. informativesresultantd'untableaudedonneesbinaires.math.sci.humaines andmsdos) E.Wol(Eds.):BeitragezurBegrisanalyse.B.I.-Wissenschaftsverlag,Mannheim,Wien,Zurich,241{25grisverbandausgewahlterraphen.Mittelseminar,THDarmstadt Duquenne,V.anduigues,J.-L.(1986):Famillesminimalesd'implications 95,5{18 Ehrenberger,P.,Heiss,R.,Ihringer,Cl.,andVogel,N.(1992):Be- Wille,R.(1982):Restructuringlatticetheory:Anapproachbasedonhierarchiesofconcepts.In:I.Rival(Ed.):Orderedsets.Reidel,Dordrecht{Boston, anter,b.andwille,r.(1995):formalebegrisanalyse:mathematische acquisitioninformalconceptanalysis.(inpreparation) Stumme,.(1995):DistributiveConceptExploration atoolforknowledge rundlagen.springer,berlin,heidelberg(toappear) 445{470 In:E.Diday(Ed.):Dataanalysis,learningsymbolicandnumericknowledge. NovaSciencePublisher,NewYork,Budapest,365{380 Wilson,R.J.(1975):Introductiontographtheory.Longman,London Wille,R.(1989):Knowledgeacquisitionbymethodsofformalconceptanalysis.