Topic Maps for Association Rule Mining

TopicMapsforAssociation Rule MiningTomášKliegr, Jan Zemánek, Marek OvečkaDepartment ofInformationandKnowledgeEngineeringFacultyofInformaticsandStatisticsUniversity ofEconomics, Prague

Data Mining using CRISP-DMThe goal of data mining is to obtain useful non-trivial patterns from the data.Analytical Report

Common data mining tasksSex(M) andSalary(Low) andDistrict(Havlickuv Brod) => Quality(Bad)Association rulesClusteringClassification

Association Rule MiningEXAMPLEUnlike clustering and classification, association rules provide true “nuggets” – rules meeting selectedinterestmeasuresDuration(2y+)andDistrict(Prague)=>Loan Quality(good)THE QUEST FOR TOPIC MAPSAntecedentConsequentTHE PROBLEM WITH INTEREST MEASURESItisusually not possible to tweaktheinterestmeasurethresholdssothatonlythereallyinterestingrules are output. To be on the safe side, we often get (many!) more rulesthandesired, Selectthereallyinterestingrulesfromtherulesoutputautomatically.Help searchingthroughtheresults.

ThequestMore precise tasks orAutomatic rule filteringThe lingua franca for exchange of data mining models is PMML

Predictive Modeling Markup LanguageXML SchemaPMML is the leading standard for statistical and data mining modelsSupported by over 20 vendors and organizationsCovers the technical part of the CRISP-DM Cyclehttp://www.dmg.org/pmml_examples/index.html

PMML is “just” an XML SchemaDeveloped for deploying mining models Good for migration from one data mining environment to anotherBut:No explicit links between nodesVerboseSelf-contained. Lacks support forInterlinking multiple PMML documentsInterlinking PMML with other information

Association Rule Mining OntologyThe ontology is a „semantization“ of PMML XML SchemaDESIGN GUIDELINESThekey design principlewas to alloweasytransformationof data from PMML to AROnSCOPEThe ontology is limited to thesubsetof PMML relevant toassociation rule mining. 60 topictypes, 50 associationtypesand 20 occurencetypesUSENo automatictransformationisyetavailable, butwe are working on oneusing OKS framework. Currently, data canbe input usingOntopoly.

xs:element ismapped to topic typeTopics are assignedsamenames as PMML NodesButrespectingspacesbetweenwordsandcapitalizationSuperclasses are introducedforsemanticallysimilar XML NodesNamedelementsused as children in otherelementsthatcarry most ofthesemanticsoftheirparents are mergedwithparentIfan XML element has a directlycorrespondingtopic type in the ontology, the URI ofthe XML element withintheschemaisused as subjectidentifierDesign guidelines: Elements

Design guidelines: AttributesEnumerationrestriction on anattributeismapped as a topic type withenumerationsuperclass (thisis a workaroundformissing TMCL support in OKS)Attributesthatcouldbeinterpreted as reference to otherelementsbecomeassociationsOtherattributesbecomeoccurencetypes

Design guidelines: AssociationsNames for association types are arbitrarily chosen so that they are most descriptiveIntroduce less rather than more associations minimizes the effort when populating the ontology from PMMLAvoid unnecessary inflation of the topic mapLink only the semantically closest topicsAdditional „soft“relations can be introduced with inference statementsorderivedwithtolog

Design guidelines: Role typesTopictypesused to map PMML elements are used as role typesUnless multiple topics are permitted in associationend. In that case superclassisused as a role type, or a new role type isintroduced

Twoalternativeassociation rulerepresentationsAprioribased(Item-Itemset)GUHA based(BooleanAttributes)

OngoingworkSupport for background knowledge „alreadyknownassociationrules“Support forschemamapping „linkingof background knowledgewithminingresults“Already in the ontology, distinguished by base ofsubjectidentifierSchemaMappinghttp://keg.vse.cz/sma/XXXBackground Knowledgehttp://keg.vse.cz/bko/xxx

Data Mining Use casePREDICT LOAN QUALITYFindclientcharacteristicsthatcouldbeused to predicttheirattitude to payingback a loan.BASED ON PAST RECORDS Input data: records on alreadygivenloans

The data6181 clients in the PKDD’99 financial datasetData were preprocessed, i.e.

….And perhaps 9997 otherassociationrulesPreprocessed dataAssociation Rule Learner

WE CAN’T PRESENT ALL 10.000 RULES TO THE CLIENTASK CLIENT WHAT HE KNOWSIfloandurationis more thantwoyearsandtheloanwasgiven in Praguedistrict, wecanexpectgoodloanquality. …background knowledge

Formalize Background Knowledge

SchemaMappingBackground knowledge can use different “vocabulary” than the data If we are to use background knowledge in querying, we need to interlink them with data.The same approach would apply if we interlink several mining models (PMMLs)

DeletinginformationwithTopicMapsFind association rules that subsume background knowledgeVisualizationof a tologquery

SummaryMethodology for transferring XML Schema to Topic MapsAssociation Rule Mining Ontology based on PMMLEasily extensible to other data mining algorithmsInitial attempts to formalize background knowledgeInitial attempts to use Topic Maps for schema mappingAROn On-Line: http://maiana.topicmapslab.de/u/lmaicher/tm/kliegr

Topic Maps for Association Rule Mining

More Related Content

Viewers also liked (13)

Similar to Topic Maps for Association Rule Mining (20)

More from tmra (20)

Recently uploaded (20)

Topic Maps for Association Rule Mining