CrowdsourcedClusteringofComputer GeneratedFloorPlans
DavidSousa-Rodrigues1 ,MafaldaTeixeiradeSampayo2(B) , Eug´enioRodrigues3 ,Ad´elioRodriguesGaspar3 ,and ´ AlvaroGomes4
1 FacultyofMaths,ComputingandTechnology,CentreofComplexityandDesign, TheOpenUniversity,MiltonKeynes,UK david.rodrigues@open.ac.uk
2 DepartmentofArchitecture,CIES,LisbonUniversityInstitute,Lisbon,Portugal mafalda.sampaio@iscte.pt
3 DepartmentofMechanicalEngineering,ADAI,LAETA, UniversityofCoimbra,Coimbra,Portugal
4 DepartmentofElectricalandComputerEngineering,INESCCoimbra, UniversityofCoimbra,Coimbra,Portugal
Abstract. Thispaperidentifiesthemaincriteriausedbyarchitecture specialistsinthetaskofclusteringalternativefloorplandesigns.Itshows howcollectiveactionsofrespondentsleadtotheirclusteringbycarrying outanonlineexercise.Thedesignswererandomlypre-generatedbya hybridevolutionaryalgorithmandaquestionnairewasposedintheend fortherespondentstoindicatewhichsimilaritycriteriatheyhaveused. Anetworkofdesignswasthenobtainedanditwaspartitionedintoclustersusingamodularityoptimizationalgorithm.Theresultsshowthat themaincriterionusedwastheinternalarrangementofspaces,followed byoverallshapeandbyexternalopeningsorientation.Thisworkallows thefuturedevelopmentofnovelalgorithmsforautomaticclassification, clustering,anddatabaseretrievalofarchitecturalfloorplans.
Keywords: Architecture · Networktheory · Crowdsourcing · Clustering · Floorplandesign · Onlinesurvey
1Introduction
Withtheadventofcomputer-aideddesign,thecomputerhasbecomemorethana meredrawingtoolorstructuralpropertiescalculator.Italsotakespartinassistingbuildingpractitionersduringthecreativeprocess,byallowingtheexploration ofpotentialsolutionsinthedailyarchitecturalpractice.InSect. 2 abriefreview ofhowthefieldisengagingwiththesenewtoolsispresented,namelyhowalgorithmsareusedtogeneratedesigns,toclassify,andmaybeusedtoretrieve architecturaldocumentsfromdatabases.Thedevelopmentofsuchalgorithms requiresaprofoundknowledgeofthewayhumanpractitionersofarchitecture perceive,understand,groupandclassifythosesamedocuments.Theaimofthis
2D.Sousa-Rodriguesetal.
studyistounderstandwhatarethecollectiveactionsofarchitecturepractitionerswhengroupingfloorplandesigns.Tothiseffectanonlinesurveywas conductedinwhichparticipantswereaskedtoselectsimilarfloorplandesigns andtoansweraquestionnaireindicatingthesimilaritycriteriaused.Theresultinganswersweremappedtoanetworkoffloorplandesignsco-selectionand wereclusteredbyamodularityoptimizationalgorithm(Sect. 3).Thefindingsof thisstudy(Sect. 4)canhelpinthedevelopmentofquerymechanismsfordatabaseretrievaloffloorplansandtheimplementationofclusteringmechanisms toaggregateresultsfromgenerativedesignmethods.Besidestheseapplications, theunderstandingofhowarchitecturepractitionerssolvethiscomplexproblem mayhelptodevelopspecificprogramsfortheteachingofarchitecture.Thelimitationandimplicationsofthisworkarebroadandrangefromthepedagogic leveltothedevelopmentofnewalgorithmsanddatabases(Sect. 5).
2RelatedWork
Oneoftheearlyarchitecturaltasksinthebuildingdesignprocessisspace planning.Architectsseektoaccommodateallrequirementsandpreferences intoarchitecturalfloorplansduringthesynthesisphase,whicharedetermined duringtheanalyticalphase.Thisisatime-consumingtrial-and-errorprocess withitsassociatedcosts.Theresultingdesignismuchdependentonthepast experienceofthearchitectandoftenbasedonalreadybuiltexamples.Asthe rooms’configurationisessentiallyacombinatorialproblem,mediumtolarge designprograms—listoffunctionalspaces,topologicalrelations,andgeometricconstraints—caneasilyreachanumberofalternativedesignsolutionsthat areimpossibletobedraftedbyhumansinthetraditionalway.Forthisreason,since1960sresearchershavebeendevelopingcomputer-basedapproaches tohelppractitioners[1, 2].Theseapproacheshavetriedtoresolvespecificdesign problemssuchasareaassignment[3],partitioningofaboundary[4 6],allocationofrooms[6 10],designadaptation[11],orthehierarchicalconstructionof differentelements[12].Iftheearlierapproacheslookedtoenumerateallpossibleconfigurations,whichledresearcherstofacethecumbersomeproblemofthe exponentialgrowthofpossiblesolutionsfordesignprogramswithmorethan8 spaces,recentstudiestriedtofindonlythemostpromisingsolutions.Toachieve this,evolutionarycomputationtechniqueswereused,asthesehaveshowncapabilitiestodealwithill-definedandcomplexproblems,anddemonstratedtoproducesurprisinglynovelsolutionsappliedtothegenerationofarchitecturalfloor plans[13].
Onlinesurveyshavebeenusedinseveralapplicationsandareamethodof datacollectionthatconveysseveraladvantages,namelytheyprovideaccessto manyindividualswhosharespecificinterestsandprofessionsthatwouldotherwisebedifficulttocontact.Surveysalsosavetimeastheydoautomated collectionofresponsesandallowresearcherstoworkonothertaskswhiledatais beingcollected[14].Whenproperlydevelopedandimplemented,asurveyportraysthecharacteristicsoflargegroupsofrespondentsonaspecifictopicand
allowsassessingrepresentativeness[15].Severaltypesofsurveysareavailable; e.g.questionnaireandinterviewformats,phonesurvey,andonlinesurveys,which canbecoupledwithinferenceenginesthatactanddirectthesurveyaccording torespondents’answers[16, 17].
Theuseofsurveysinarchitecturalenvironmentshasbeenconductedinmany aspectsofthediscipline.Theyhavebeenusedintheestablishinggroundtruths inperceptualunderstandingoffloorplansforthecharacterizationofshapes,lines andtexture[18, 19].Feedbacktoarchitectsisbeinggivenbysurveysofarchitecturevirtualimmersiveexperimentsthataimtounderstandphysiologicalsignals ofemotions,namelyfear,inspaceperception[20, 21].Severalstudieshavebeen proposedthatincludetheparticipationofthecrowdandarebottom-uplearningprocesses,e.g.peerassessment[22]wherestudentsmarkeachother’work. Inthisstudytheprocessofgroupingfloorplansisinvestigatedtounderstand thecriteriausedbythestudentsandotherpractitionersduringthegrouping process.
3MethodsandMaterials
3.1TheOnlineSurvey
Anonlinesurveywassetupasanexercisetocollectinformationonhowthe respondentsperformtheclusteringtask.Therearemanyonlinetoolsforconductingsurveys[14, 15],butnonecanhandlethespecialproblemposedbyusing architecturaldocuments.Therefore,itwasdecidedtodevelopawebapplication fortheexperiment(Fig. 1).Tounderstandtheperceptionandcriteriaofthetargetpopulation,apostexperimentquestionnairewaspresentedtoparticipants. Individualswhosedailyactivitiesarerelatedtobuildingdesignwerechosenas thetargetgroup;i.e.architects,architecturestudents,civilengineers,andurban planners.Asthetargetgroupiswelldelimited,theselectionofparticipantswas conductedthroughUniversitycommunitiesofarchitecturestudentsandformer students,andalsothroughtheprofessionalaffiliationcontactlists.Thisensured thatthemajorityoftheparticipantsinthisstudywererelatedtothesubject oriftheirpresentprofessionaloccupationisnotrelatedtoarchitecture,atleast theyreceivedtraininginarchitecture.
Fromatotalof72generatedfloorplans,twelvewererandomlyselectedand displayedinawebinterface.Theuserwasaskedtodrag-and-droptoaspecific areainthescreenthefloorplansthatheconsideredsimilar.Eachrespondent repeatedthisiterationtentimes.The72floorplansweregeneratedusingthe EvolutionaryProgramfortheSpaceAllocationProblem(EPSAPalgorithm) [7 9].Thisalgorithmiscapableofproducingalternativefloorplansaccordingto theuser’spreferencesandrequirementssetasthebuildingfunctionalprogram. Thesolutionsgeneratedwereforasingle-familyhousewiththreebedrooms,one hall,onekitchen,alivingroom,onecorridorandtwobathrooms.Onebathroom andallbedroomsareconnectedtothecorridor.Theremainingspacesareconnectedtothehall.Thekitchenpresentsaninternaldoorconnectingittothe livingroom.Oneofthebathroomsservesthepublicareasofthehouse,whilethe
Fig.1. Usersdrag-and-dropfloorplansintotheshadedareaaccordingtosimilarity
otherconnectstothecorridoroftheprivateareaofthehouse.Allinnerrooms havedoorsof90cmwidth,theexceptionbeingthelivingroomdoorsthatare 140cm.Exceptforthecirculationareasandoneofthebathrooms;allareas haveatleastonewindow.Thehallhasanexteriordoorfacingnorth.Noother restrictionswereimposed.Attheendoftheonlinesurvey,afinalquestionnaire waspresentedwithalistofpossiblecriteriaandtheusercouldchosewhichhe usedortoprovidewrittenalternatives.Aftersubmissiontheexerciseendedand theuserwasredirectedtothehomepage.
3.2NetworkScienceAnalysisoftheCollectedData
Anormalizedmatrixdepictingthefractiontotimeseachpairoffloorplanswas co-selectedisconstructed.Thismatrixisunderstoodasanadjacencymatrix wheretheentriesrepresenttheweightsoftheconnectionsbetweentwodesigns. Theresultspresentsomebackgrounduncertaintyanditisnecessarytodefine aminimumthresholdfortheentriesofthematrix.Thethresholdvaluewas testedtoidentifythestructureoftheselectionprocess,whichrepresentedthe floorplan’snetwork.Thisnetwork—undirectedandweighted—ispartitioned withtheedgebetweennesscommunitydetectionalgorithm[23].Thisisadivisive hierarchicalmechanismthataimstofindcommunitiesbymaximizingthevalueof modularity—networkswithhighmodularityhavedenseintraclusterconnections butsparseconnectionsbetweenverticesofdifferentclusters.Thealgorithmfor creatingthedendrogramproceedsinthefollowingmanner[23]:
1.Calculatetheedgebetweennessinthenetwork.
2.Removetheedgewithhighestvalueofbetweenness.
3.Recalculatebetweennessforalledgesaffectedbytheremoval.
4.Repeatfromstep2untilallnodesareisolated(noedgesremain).
Thebetweennesscentralityofanedgeisthesumofthefractionofall-pairs shortestpathsthatpassthroughthatedge[24 26].Thegraphandtheresulting partitionarecharacterizedaccordingtodiverseproperties—averagepathlength, density,andclusteringcoefficient.
4Results
Atotalof609invitationstoparticipateintheonlinesurveyweresubmitted. Thesurveywasavailablefortherespondentsduringtwoweeks.Ofthoseinvitations,202personsansweredthesurveybyreadingtheinformedconsent,filling theoptionaldemographicinformationformandinitiatedtheexperiment.Of those202only110carriedoutthe10iterationsaskedandfilledthefinalcriteria questionnaire.Intotal,therespondentsperformed1257iterations.Oftheparticipantsthatregistered,92didnotconcludetheexercise.Theaveragenumber ofiterationsmadebythose92personswas1,7.Thepoolofparticipantsinhabits mainlyinPortugalandtheagesrangebetween18and50yearsold.
Byvaryingthethresholdofthefractionofco-selectionsoffloorplandesigns, itispossibletoverifythattheinitialdensenetworkpresentslowmodularity,high densityofedges,andsmallaveragepathlength(Fig. 2).Italsopresentsahigh clusteringcoefficient,whichisindicativeofmanytrianglesinthenetwork.The
andaveragepathlength(rightaxis).
Fig.3. Dendrogramoftheclusteringidentifyingtheresultingclusters.
increaseofthethresholdleadstohigheraveragepathlength,loweredgedensity, andlowerclusteringcoefficient.Modularitystartsrisinguntilathresholdof0,36. Theaveragepathlengthpeaksaroundathresholdof0,19andavalueof4,6and afterstartsdecreasingagainasaresultofthefragmentationoftheresulting networkweremanyisolatednodesemerge.
Duetothisfragmentation,athresholdof12%waschosen.Thisstillensured amodularityvalueof0,44andanaveragepathlengthof2,7.Theclustering resultedin13clusters(Fig. 3).Itisclearfromthedendrogramthatthisclustering presentsomeisolatednodes,namely {1, 25, 46}.Thedistributionofthefloorplan designclusterswas {20, 18, 7, 5, 5, 3, 3, 3, 3, 2, 1, 1, 1}.
Theclusterspresentgoodinternalconsistency,meaningthatuponinspection theyarecoherentwiththecriteriareportedbytheusers.Thiscanbeseeninthe threeexamplesoftheclustersobtainedfromtheclusteringprocessinFigs. 4, 5, and 6
Thesurveyco-criteriaanalysisindicatesthattwocriteriaareoftenselected togetherbyparticipants, byinteriorspaces and bycirculationspaces (Table 1). Onasecondlevel,theparticipantsconsideredcriteriarelatedtotheoverall shape
Fig.4. Clusterofplans {0, 13, 41, 63, 64}
Fig.5. Clusterofplans {3, 7, 50}
Fig.6. Clusterofplans {4, 9, 10, 11, 19, 42, 48}
Table1. Criteriaco-selection.Diagonalentriesrepresentfrequencyofeachcriterion. Intersectionsofthelowertriangleindicatefrequencyofco-selectionoftwocriteria.
(eitherconsideringcaseswheremirroringorrotationsoccurred)andinathird tierrespondentsconsideredtheexistenceof externalopenings .Thisclearlyshows thatusersfavortheinteriorspaceorganizationasthemostimportantfeaturein definingsimilarityoffloorplans.
5Conclusion
Theresults,asshowninTable 1,indicatethatarchitecturepractitionersgive higherimportancetotheinteriorconfigurationsofspacesthantheoverallbuildingshape.ThisinformationisimportantforfuturedevelopmentofICT-mediated strategiesforarchitectureeducationandprofessionalpractitioners.Theywill alsoimpactotherapplicationssuchasfloorplandesigndatabaseretrieval—by identifyingtheencodingfeaturesusedbyhumanpractitionersthatcanthenbe implementedintheencodingofdatabaserecords—andaggregationofsimilar solutionsthatresultfromgenerativedesignmethods—releasinghumansfrom
8D.Sousa-Rodriguesetal.
thetediousandrepetitivetaskofgroupingsimilarfloorplans,andallowingfor concisetypologicalpresentationoffloorplansinautomatedways.
Theexecutionofonlinesurveysisnotfreeofproblems.Samplingissues mightbepresent,astherespondentsarenotmonitoredandsomemisbehavior canhappen;e.g.doubleanswering[14].Theminimizationoftheseproblemswas achievedbyassigningauniquefive-digitcodetoeachparticipantthatmatches theanswersinthedatasetwiththeIPaddress.Theproblem“lurkers”was minimizedbycontactingeachparticipantdirectly,thatispeoplewhodonot participatebuthaveaccesstothesurvey[14].
Theresponseratewasaround30%.Althoughmanyonlinesurveyshavelow responserates,theytrytoincreasebysomeincentivemechanisms,e.g.financial incentives,prizes,coupons,orbooks.However,inthiscasethatwasnotanissue. Noincentivemechanismwasimplementedinthisexperimentforthecompletion ofthesurvey.Thepersonalcontactoftheresearcherwitheachparticipantmade theparticipationinthesurveyamatterofpersonalandprofessionalrespect. However,theeffectivecompletionratewassmall,as92participantsdidnot completethesurvey(18%).Theselimitationsarenotexclusivetothiskindof onlinesurveytechnique[14, 15].
Theseresults,namelythecriteriareportedbytherespondents,canbeincorporatedinmachinelearningalgorithmstoperformclusteringtasksinwaysthat mimicexperts’actions.Also,theobtainedclusteringresultswillbeusedasa groundtruthorbenchmarkfornewclusteringalgorithmsthatdealwithperceptualclusteringoffloorplandesigns.
Acknowledgements. Sousa-Rodrigues,D.waspartiallysupportedbyprojectTopdrimFP7-ICT-2011-8/318121.Rodrigues,E.,Gaspar,A.R.,andGomes, ´ A.werepartiallysupportedbyproject AutomaticGenerationofArchitecturalFloorPlanswith EnergyOptimization (GerAPlanO),QREN38922,CENTRO-07-0402-FEDER-038922 andframedunderthe EnergyforSustainabilityInitiative atUniversityofCoimbra.
References
1.Liggett,R.S.:Automatedfacilitieslayout:past,presentandfuture.Autom.Constr. 9(2),197–215(2000)
2.Steadman,P.:Generativedesignmethodsandtheexplorationofworldsofformal possibility.Architect.Des. 84(5),24–31(2014)
3.Gero,J.S.,Kazakov,V.A.:Evolvingdesigngenesinspacelayoutplanningproblems.Artif.Intell.Eng. 12(3),163–176(1998)
4.Mitchell,W.J.,Steadman,J.P.,Liggett,R.S.:Synthesisandoptimizationofsmall rectangularfloorplans.Environ.Plan.B 3(1),37–70(1976)
5.Quiroz,J.C.,Louis,S.J.,Banerjee,A.,Dascalu,S.M.:Towardscreativedesign usingcollaborativeinteractivegeneticalgorithms.In:EvolutionaryComputation 2009,pp.1849–1856.IEEE,May2009
6.Koenig,R.,Knecht,K.:Comparingtwoevolutionaryalgorithmbasedmethodsfor layoutgeneration:densepackingversussubdivision.Artif.Intell.Eng.Des.Anal. Manuf. 28(03),285–299(2014)
7.Rodrigues,E.,Gaspar,A.,Gomes, ´ A.:Anevolutionarystrategyenhancedwith alocalsearchtechniqueforthespaceallocationprobleminarchitecture,part1: methodology.Comput.Aided-Des. 45(5),887–897(2013)
8.Rodrigues,E.,Gaspar,A.,Gomes, ´ A.:Anevolutionarystrategyenhancedwith alocalsearchtechniqueforthespaceallocationprobleminarchitecture,part2: validationandperformancetests.Comput.Aided-Des. 45(5),898–910(2013)
9.Rodrigues,E.,Gaspar,A.,Gomes, ´ A.:Anapproachtothemulti-levelspaceallocationprobleminarchitectureusingahybridevolutionarytechnique.Autom.Constr. 35,482–498(2013)
10.Michalek,J.J.,Choudhary,R.,Papalambros,P.Y.:Architecturallayoutdesign optimization.Eng.Optim. 34,461–484(2002)
11.Garza,A.G.D.S.,Maher,M.L.:GENCAD:ahybridanalogical/evolutionarymodel ofcreativedesign.In:Gero,J.S.,Maher,M.L.(eds.)ComputationalAndCognitive ModelsOfCreativeDesignV,pp.141–171.KeyCenterofDesignComputingand Cognition,UniversityofSidney,Sydney(2001)
12.Virirakis,L.:GENETICA:acomputerlanguagethatsupportsgeneralformal expressionwithevolvingdatastructures.IEEETrans.Evol.Comput. 7(5),456–481(2003)
13.Kalay,Y.E.:Architecture’sNewMedia:Principles,TheoriesandMethodsof Computer-AidedDesign.TheMITPress,Cambridge(2004)
14.Wright,K.B.:Researchinginternet-basedpopulations:advantagesanddisadvantagesofonlinesurveyresearch,onlinequestionnaireauthoringsoftwarepackages, andwebsurveyservices.J.Comput.-MediatedCommun. 10(3),00–00(2005)
15.Evans,J.R.,Mathur,A.:Thevalueofonlinesurveys.InternetRes. 15(2),195–219 (2005)
16.Urbano,P.,Sousa-Rodrigues,D.:Theadvantageofusingrulesinonlinesurveys. RevistadeCiˆenciasdaComputa¸c˜ao III(3),38–49(2008)
17.Urbano,P.,Sousa-Rodrigues,D.:Rulebasedsystemsappliedtoonlinesurveys. In:IADISWWW/InternetConference.Freiburg,Oct2008
18.Sousa-Rodrigues,D.,deSampayo,M.T.,Rodrigues,E.,Gaspar,A.R.,Gomes, ´ A.,Antunes,C.H.:Onlinesurveyforcollectiveclusteringofcomputergenerated architecturalfloorplans.arXivpreprint arXiv:1504.08145 (2015)
19.delasHeras,L.P.,Fernandez,D.,Fornes,A.,Valveny,E.,Sanchez,G.,Llados, J.:Perceptualretrievalofarchitecturalfloorplans.In:10thIAPRInternational WorkshoponGraphicsRecognition(2013)
20.Dias,M.S.,Eloy,S.,Carreiro,M.,Proˆen¸ca,P.,Moural,A.,Pedro,T.,Freitas,J., Vilar,E.,d’Alpuim,J.,Azevedo,S.:Designingbetterspacesforpeople:virtual realityandbiometricsensingastoolstoevaluatespaceuse.In:Gu,N.,Watanabe, S.,Erhan,H.,Haeusler,M.H.,Huang,W.,Sosa,R.(eds.)RethinkingComprehensiveDesign:SpeculativeCounterculture,Proceedingsofthe19thInternational ConferenceoftheAssociationofComputer-AidedArchitecturalDesignResearch inAsia,HongKong(2014)
21.Dias,M.S.,Eloy,S.,Carreiro,M.,Vilar,E.,Marques,S.,Moural,A.,Proˆen¸ca,P., Cruz,J.,d’Alpuim,J.,Carvalho,N.,Azevedo,A.S.,Pedro,T.:Spaceperception invirtualenvironments-onhowbiometricsensinginvirtualenvironmentsmay givearchitectsusers’sfeedback.In:Thompson,E.M.(ed.)Fusion-Proceedings ofthe32ndeCAADeConference,NewcastleuponTyne,England,UK,vol.2,pp. 271–280,September2014
22.deSampayo,M.T.,Sousa-Rodrigues,D.,Jimenez-Romero,C.,Johnson,J.H.:Peer assessmentinarchitectureeducation.In:InternationalConferenceonTechnology andInnovation,Brno,CzechRepublic,September2014
23.Girvan,M.,Newman,M.:Communitystructureinsocialandbiologicalnetworks. Proc.Natl.Acad.Sci.USA 99,7821–7826(2002)
24.Anthonisse,J.M.:Therushinadirectedgraph.Technicalreport,StichtingMathematicshCentrum,Amsterdam(1971)
25.Freeman,L.C.:Asetofmeasuresofcentralitybasedonbetweenness.Sociometry 40(1),35–41(1977)
26.Brandes,U.:Afasteralgorithmforbetweennesscentrality*.J.Math.Sociol. 25(2), 163–177(2001)